MOE and MOR assessment technologies for improving graded ...

PRODUCTS & PROCESSING 

PROJECT NUMBER: PNB040-0708 NOVEMBER 2009 

MOE and MOR assessment 

technologies for improving 

graded recovery of exotic 

pines in Australia 

This report can also be viewed on the FWPA website 

www.fwpa.com.au 

FWPA Level 4, 10-16 Queen Street, 

Melbourne VIC 3000, Australia 

T +61 (0)3 9927 3200 F +61 (0)3 9927 3288 

E info@fwpa.com.au W www.fwpa.com.au 

`

MOE and MOR assessment technologies for 

improving graded recovery of exotic pines in 

Australia 

Prepared for 

Forest & Wood Products Australia 

by 

H. Bailleres, G. Hopewell and G. Boughton

Publication: MOE and MOR assessment technologies for 

improving graded recovery of exotic pines in Australia 

Project No: PNB040-0708 

© 2009 Forest & Wood Products Australia Limited. All rights reserved. 

Forest & Wood Products Australia Limited (FWPA) makes no warranties or assurances with 

respect to this publication including merchantability, fitness for purpose or otherwise. FWPA and 

all persons associated with it exclude all liability (including liability for negligence) in relation to 

any opinion, advice or information contained in this publication or for any consequences arising 

from the use of such opinion, advice or information. 

This work is copyright and protected under the Copyright Act 1968 (Cth). All material except the 

FWPA logo may be reproduced in whole or in part, provided that it is not sold or used for 

commercial benefit and its source (Forest & Wood Products Australia Limited) is acknowledged. 

Reproduction or copying for other purposes, which is strictly reserved only for the owner or 

licensee of copyright under the Copyright Act, is prohibited without the prior written consent of 

Forest & Wood Products Australia Limited. 

ISBN: 978-1-920883-89-8 

Researcher: 

H. Bailleres, G. Hopewell 

Queensland Primary Industries and Fisheries 

Gate 3/ 80 Meiers Rd 

Indooroopilly Q 4068 

Subcontractors: 

G. Boughton 

TimberED Services Pty Ltd 

PO Box 30 

Duncraig East WA 6023 

L. Branchiau, 

CIRAD Xylometry 

73, Rue J.F. Breton, BP 5035 

34032 Montpellier France 

Final report received by FWPA in November, 2009 

Forest & Wood Products Australia Limited 

Level 4, 10-16 Queen St, Melbourne, Victoria, 3000 

T +61 3 9927 3200 F +61 3 9927 3288 

E info@fwpa.com.au 

W www.fwpa.com.au

Executive Summary 

This project was designed to provide the structural softwood processing industry with the 

basis for improved green and dry grading to allow maximise MGP grade yields, consistent 

product performance and reduced processing costs. To achieve this, advanced statistical 

techniques were used in conjunction with state-of-the-art property measurement systems. 

Specifically, the project aimed to make two significant steps forward for the Australian 

structural softwood industry: 

• assessment of technologies, both existing and novel, that may lead to selection of a 

consistent, reliable and accurate device for the log yard and green mill. The purpose is 

to more accurately identify and reject material that will not make a minimum grade of 

MGP10 downstream; 

• improved correlation of grading MOE and MOR parameters in the dry mill using new 

analytical methods and a combination of devices. 

The three populations tested were stiffness-limited radiata pine, strength-limited radiata pine 

and Caribbean pine. Resonance tests were conducted on logs prior to sawmilling, and on 

boards. Raw data from existing in-line systems were captured for the green and dry boards. 

The dataset was analysed using classical and advanced statistical tools to provide correlations 

between data sets and to develop efficient strength and stiffness prediction equations. 

Stiffness and strength prediction algorithms were developed from raw and combined 

parameters. 

Parameters were analysed for comparison of prediction capabilities using in-line parameters, 

off-line parameters and a combination of in-line and off-line parameters. 

The results show that acoustic resonance techniques have potential for log assessment, to sort 

for low stiffness and/or low strength, depending on the resource. From the log measurements, 

a strong correlation was found between the average static MOE of the dried boards within a 

log and the predicted value. These results have application in segregating logs into structural 

and non-structural uses. Some commercial technologies are already available for this 

application such as Hitman LG640. 

For green boards it was found that in-line and laboratory acoustic devices can provide a good 

prediction of dry static MOE and moderate prediction for MOR.There is high potential for 

segregating boards at this stage of processing. Grading after the log breakdown can improve 

significantly the effectiveness of the mill. Subsequently, reductions in non-structural volumes 

can be achieved. Depending on the resource it can be expected that a 5 to 8 % reduction in 

non structural boards won’t be dried with an associated saving of $70 to 85/m 3 

For dry boards, vibration and a standard Metriguard CLT/HCLT provided a similar level of 

prediction on stiffness limited resource. However, Metriguard provides a better strength 

prediction in strength limited resources (due to this equipment’s ability to measure local 

characteristics). The combination of grading equipment specifically for stiffness related 

predictors (Metriguard or vibration) with defect detection systems (optical or X-ray scanner) 

provides a higher level of prediction, especially for MOR. Several commercial technologies 

are already available for acoustic grading on board such those from Microtec, Luxscan, 

Falcon engineering or Dynalyse AB for example. 

i

Differing combinations of equipment, and their strategic location within the processing chain, 

can dramatically improve the efficiency of the mill, the level of which will vary depending of 

the resource. For example, an initial acoustic sorting on green boards combined with an 

optical scanner associated with an acoustic system for grading dry board can result in a large 

reduction of the proportion of low value low non-structural produced. 

The application of classical MLR on several predictors proved to be effective, in particular for 

MOR predictions. However, the usage of a modern statistics approach (chemometrics tools) 

such as PLS proved to be more efficient for improving the level of prediction. 

Compared to existing technologies, the results of the project indicate a good improvement 

potential for grading in the green mill, ahead of kiln drying and subsequent cost-adding 

processes. The next stage is the development and refinement of systems for this purpose. 

ii

Table of Contents 

Executive Summary .................................................................................................................... i 

Introduction ................................................................................................................................1 

Literature review ........................................................................................................................ 3 

Radiata pine................................................................................................................................3 

Caribbean pine............................................................................................................................ 4 

Non-destructive testing methods................................................................................................ 4 

Radiography ........................................................................................................................... 5 

Evaluation of visual characteristics........................................................................................ 6 

Near-infrared (NIR) methods................................................................................................. 6 

Microwave techniques............................................................................................................ 6 

Stress wave methods .............................................................................................................. 7 

Ultrasonic ........................................................................................................................... 7 

Sonic................................................................................................................................... 8 

Resonance method.............................................................................................................. 8 

Remark on the measured acoustic velocities.................................................................... 10 

Machine stress rating............................................................................................................ 10 

Static bending........................................................................................................................... 11 

NDT of wood for grading: results from previous studies .................................................... 12 

Standard requirements.............................................................................................................. 15 

Study Methodology.................................................................................................................. 16 

Materials............................................................................................................................... 16 

Sampling method.............................................................................................................. 16 

Kiln drying ....................................................................................................................... 18 

Equipment and methods....................................................................................................... 18 

Non destructive testing- mechanical properties measurement......................................... 18 

Machine stress rating............................................................................................................... 18 

In-line acoustic test .................................................................................................................. 19 

Off-line acoustic test................................................................................................................. 19 

Gamma ray............................................................................................................................... 22 

Non destructive testing- structural properties measurement............................................ 23 

Linear high grader (LHG)........................................................................................................ 23 

WoodEye ® ........................................................................................................................ 23 

Destructive standard static bending tests ......................................................................... 24 

Results ...................................................................................................................................... 29 

Static bending....................................................................................................................... 29 

Biased and random tests together..................................................................................... 29 

Random and biased .......................................................................................................... 32 

Biased vs Random............................................................................................................ 37 

Position of the boards within the logs .............................................................................. 39 

Individual NDT predictors for static MOE and MOR biased evaluation ............................ 40 

At dry mill level ............................................................................................................... 40 

Radiata E resource........................................................................................................... 40 

Radiata R resource........................................................................................................... 43 

Caribbean resource.......................................................................................................... 47 

At green mill level............................................................................................................ 50 

Radiata E resource........................................................................................................... 50 

Radiata R resource........................................................................................................... 52 

Caribbean resource.......................................................................................................... 53 

Combinations of NDT predictors for static MOE and MOR evaluation ............................. 55

Radiata E resource............................................................................................................ 55 

Radiata R resource ........................................................................................................... 57 

Caribbean resource........................................................................................................... 58 

Prediction of static bending MOE and MOR from logs....................................................... 60 

Radiata E .......................................................................................................................... 60 

Radiata R.......................................................................................................................... 60 

Caribbean ......................................................................................................................... 61 

Practical grading into MGP grades ...................................................................................... 62 

Using best vibration prediction for logs........................................................................... 62 

Using best prediction for dry boards NDT technologies...................................................... 63 

Discussion and conclusions...................................................................................................... 72 

Recommendations .................................................................................................................... 74 

References ................................................................................................................................ 75 

Internet references................................................................................................................ 77 

Standards .............................................................................................................................. 77 

Personal communication ...................................................................................................... 78 

Acknowledgements .................................................................................................................. 79 

Researcher’s Disclaimer........................................................................................................... 80 

Appendix 1. Resonance method extracted signal descriptors.................................................. 81 

Appendix 2. WoodEye profile descriptions. ............................................................................ 83 

Appendix 3. Biased and random testing protocol. ................................................................... 84 

Radiata E resource................................................................................................................ 84 

Radiata R and Caribbean resources...................................................................................... 84

Introduction 

In Australia, a range of native and exotic softwood forests and plantations provide an 

important source of fibre for sawn, round and panel products. The introduced Pinus species 

are grown in plantations managed specifically to produce high volumes of structural products 

primarily for the domestic construction sector. The native softwoods such as white cypress 

(Callitris glaucophylla Thompson & Johnson) and hoop pine (Araucaria cunninghamii Aiton 

ex D. Don) were not included in this study and all further reference to the term ‘softwood/s’ 

in this document, indicates plantation-grown exotic pine. 

The scope of work was identified as a priority by the Australian softwood sector, and is the 

first study of its type conducted in Australia. Subsets of the radiata pine and Caribbean pine 

populations represent the range of mechanical properties of the majority of the Australiangrown 

exotic pine resource. This sector requires a universally applicable method to accurately 

and rapidly predict the strength and stiffness of green logs and boards processed for structural 

products. This will allow non-structural and low stiffness boards to be diverted before 

entering the dry chain. 

There is often a high variability in wood properties of fast grown trees harvested at a young 

age, even within a single stem (Zobel and Buijtenen, 1989). Consequently, within a log or 

even within a board, the wood properties may be significantly different. Wood can be 

characterized as a highly heterogeneous and highly anisotropic material. Heterogeneous 

means that wood does not have a uniform structure and this variability can affect strength, for 

example knots, resin pockets and reaction wood. Anisotropic means that wood is a very 

oriented material, in other words directionally dependent with different properties in different 

planes. The strength and stiffness in the longitudinal direction of the tree are much higher than 

in the transverse directions. This effect can cause problems when the grain direction is not 

always parallel to the sawn direction of boards. High slope of grain can seriously decrease the 

bending strength. Because such variation is not acceptable in wood used for structural 

applications, it must be appropriately graded to ensure safety and performance in service. 

Grading is the process by which timber is sorted into appropriate stress grades with consistent 

properties in each grade. Inevitably, there is a range of properties within a grade and 

significant overlap in properties between groups. Currently, processors undertake the grading 

task in the dry mill. To prevent unnecessary processing, associated costs and energy usage, 

for example kiln drying of low strength and ultimately non-structural boards, the grading 

process and subsequent segregation should be done as early as possible in the value chain. 

This project was initiated and designed in order to address this issue. 

The main objectives of this project were: 

• To improve the use of robust predictors of strength and stiffness acquired through 

existing in-line processing equipment such as Metriguard Continuous Lumber Tester 

(CLT), knot area ratio (KAR), acoustics, gamma ray and optical scanners. 

• To improve the use of grading tools for upstream sorting: from logs, green boards and 

finally dry mill products. 

• To identify if new parameters are available to refine current predictive mechanisms, 

and determine the most effective way to input these into prediction equations. 

• To define the accuracy of the relevant parameters, or combination of parameters, for a 

number of technologies to independently improve grading systems. 

1

• To develop advanced vibrational methods in order to grade softwood boards and logs 

for structural purposes. 

The trials discussed here included three separate samples from distinct populations: stiffness 

limited radiata pine, strength-limited radiata pine and Caribbean pine. For each sample, logs 

were measured and weighed to provide density data, then tested for acoustic resonance. The 

resulting MOE calculations were later correlated with results from reference tests on the 

boards sawn from the logs. 

After sawing, the green boards were subjected to a range of tests to provide relevant data as 

summarised below: 

• Industrial acoustic test (Weyerhaeuser Thumper for MOE); 

• Gamma ray (Geological & Nuclear sciences Ltd for moisture content and density); 

• Metriguard Continuous Lumber Tester (MOE); 

• Metriguard High Capacity Lumber Tester (MOE); 

• LHG X-ray (for density and knot area ratio, strength-limited radiata pine only); 

• WoodEye ® (laser and camera optical scanner for defect type, size and position at 

production speed); 

• Acoustic measurements (Bing ® ; longitudinal and transverse MOEs). 

• Reference tests (MOE and MOR, AS 4063:1992). 

Reference tests (static bending for MOE and MOR, AS 4063:1992) including both biased and 

random samples, were undertaken on a universal static 4-point bending test machine and 

visual defects were measured. 

The dataset was analysed using mathematical and chemometrics’ statistical tools to extract 

relevant parameters and to provide correlations between data sets and develop efficient 

strength and stiffness prediction equations. Stiffness and strength prediction algorithms were 

developed from raw and combined parameters using specific chemometrics’ tools including 

multi-variate linear regression (MLR) and partial least squares (PLS). 

The parameters were analysed for comparison of prediction capabilities from in-line 

parameters, off-line parameters (vibration analysis and manual defects measurements) and a 

combination of in-line and off-line parameters. 

The predictors which provided the best correlations to the static MOE were used to sort the 

boards into MGP grades 

The results from the project will allow processors to improve existing machine grading and 

assist in the development of next generation systems for more accurate grading of structural 

wood. The improved confidence in estimating strength and stiffness values for dried timber 

will allow for grading closer to threshold limits, thus improving structural grade yields 

resulting in a more efficient and profitable use of the softwood resource. This equates to 

greater resource optimisation, reduced costs and increased profits for the softwood sector. 

2

Literature review 

Radiata pine 

Pinus radiata D. Don is marketed in Australia as radiata pine , but it is also known as 

Monterey pine and Insignis pine. It is easy to establish, can grow under a range of site 

conditions and produces large quantities of useable wood over relatively short rotations. 

Radiata pine from two disjunct sources viz south east Australia and south-west Australia, was 

investigated. Radiata pine is a versatile and widely planted species and in Australia there are 

almost 720,000 hectares (Web 1, 2003) of established radiata pine plantation forests. 

Native radiata pine occurs at five locations in central to north America. Three of these 

locations are in California: Año Nuevo, the Monterey Peninsula and Cambria. The other two 

are found on the two small islands off the coast of Mexico, Cedros (Pinus radiata var. 

cedrosensis) and Guadalupu (Pinus radiata var. binata) (Bootle, 2005). 

The wood of radiata pine is pale yellow-brown and is generally straight grained with 

prominent growth rings formed by alternating bands of earlywood and latewood. It has low 

natural durability, however may be treated with preservatives for outdoor use. The wood has a 

good strength to weight ratio, with good nail holding and gluing ability, resistance to nail 

splitting and is relatively easy to saw and dry (Bootle, 2005). 

The air-dry density (12% moisture content) is variable but typically is around 545 kg/m 3 and 

nominated strength groups are S6, SD6 (Hopewell, 2006). Wood density increases as the tree 

ages, and is influenced by environmental factors. New Zealand research has found that trees 

grown at lower altitude in warmer areas have a higher wood density than those grown at 

higher altitudes in cooler areas (Kininmonth and Whitehouse, 1991). 

The innermost annual rings in the tree stem have a different anatomical structure compared to 

the wood in outer layers of the stem. The wood in the innermost rings is known as ‘juvenile 

wood’ and it has significantly different mechanical properties than the outer ‘adult wood’. 

The central core generally exhibits pronounced spiral grain, shorter fibres and lower wood 

density (as low as 350 kg/m 3 at 12% MC). This corewood is usually confined to the first ten 

to twenty growth rings only and is regarded as low-quality wood with the following 

characteristics (Ilic et al, 2003, except as noted): 

• wide growth-rings; 

• high grain spirality; 

• low density and stiffness; 

• thin cell walls and short trachieds; 

• high longitudinal shrinkage; 

• low transversal shrinkage 

• presence of compression wood, and 

• lower cellulose:lignin ratio (Bendtsen, 1978). 

Radiata pine dries rapidly, and is usually kiln dried from the green condition at high 

temperatures e.g. 140°C. The wood is easy to dry but boards sawn from the central core zone 

which are prone to distortion. Improved stability of the seasoned product is achieved by presteaming 

for several hours and the use of concrete stack weights during drying (Bootle, 2005). 

Radiata pine has a wide range of structural and decorative uses including framing, furniture, 

paneling, lining, glued laminated beams, veneer, plywood and pulp. When treated with 

3

preservatives, it can be used for outdoor applications such as posts and poles (Bootle, 2005). 

Small diameter logs produced from thinning operations can be used for posts or pulpwood. 

Caribbean pine 

Caribbean pine sourced from south-east Queensland provided the balance of test material for 

the trials. Caribbean pine has been planted in Queensland and New South Wales where it has 

developed a reputation for excellent growth with minimal branching, even on poorly drained 

soils, resulting in desirable wood quality viz smaller and less frequent knots than related Pinus 

species. 

Caribbean pine is pale yellow to yellow for the sapwood, and yellow to pale brown, with pink 

tints for the heartwood. The difference of colour between earlywood and latewood is 

pronounced, and the grain is straight with a coarse and uneven texture. 

It has a low natural durability but can be effectively impregnated with chemical preservatives. 

The air-dry density ranges from 545 to 575 kg/m³ (Hopewell, 2006). Provisional strength 

groups have been assigned to Caribbean pine as (S6) and (SD6). 

Caribbean pine dries rapidly, and is usually kiln dried from the green condition at high to very 

high temperatures e.g.140-180°C without loss of strength (Siemon, 1981). The wood is easy 

to dry, except boards sawn from the central core zone which are prone to distortion. Improved 

stability of the seasoned product is achieved by reconditioning and the use of stack weights 

during drying. 

Caribbean pine has a wide range of uses including framing, flooring, mouldings, joinery, 

furniture, plywood, treated landscaping and roundwood products, laminated beams, medium 

density fibreboard and paper production. Resin can be a problem during sawmilling as it 

builds up on saws and other processing equipment (Bootle, 2005). 

Non-destructive testing methods 

Non-destructive testing (NDT), also called non-destructive evaluation (NDE) and nondestructive 

inspection (NDI), is the science of identifying physical and mechanical properties 

of a piece of material without altering its end-use capabilities (Ross and Pellerin, 1994). 

Since the 1920’s, NDT methods have developed from laboratory testing to an indispensable 

production tool. Often components are too costly, or destructive testing is not possible, thus 

NDT is becoming increasingly important as a quality control management tool in almost 

every manufacturing process. 

Today there are a large variety of NDT methods which are used worldwide to detect material 

characteristics such as: variation in structure, the presence of cracks or other physical or 

mechanical discontinuities, dimensions of products and to determine other characteristics of 

material. The most common methods are listed below (Bucur, 2003): 

• visual and optical testing; 

• ultrasonic testing; 

• radiographic testing; 

• impulse excitation technique; 

• electromagnetic testing; 

• vibration methods; 

4

• magnetic resonance; 

• laser testing; 

• near-infrared. 

Many of these methods originate from the medical field and have been adapted for industrial 

use (for example computed radiography and ultrasonic). These and other techniques are used 

in a wide range of industrial activities including aviation, aerospace, construction, 

manufacturing, pipelines and railways. Each method has advantages and disadvantages, for 

example radiographic methods can be used on several materials to detect the internal 

condition and/or properties of samples but is limited by the thickness of the test material and 

requires high safety precautions. Some methods are more suitable for detection and evaluation 

of certain flaws/properties than others. Therefore, the right choice or the combination of 

methods is desirable. 

Using NDT to evaluate the physical properties of wood has its’ origin in the need to solve 

practical problems without destruction of the integrity of the object under inspection (Bucur, 

2003). The heterogeneity and anisotropy of wood make it difficult for manufacturers of forest 

products to provide a consistent quality to their customers. Many of the methods listed above 

have been adapted for predicting the performance of wood, but its wide variability makes it 

more challenging than for homogenous materials like metal and plastics (Ross and Pellerin, 

1994). Therefore, research work on NDT of wood is required to determine a more accurate 

performance of a wood member (Ibid). Among the wood characteristics to be assessed nondestructively, 

strength and stiffness are the most important for structural applications. The 

only way to determine the true strength of a piece of timber is to break it. But afterwards it is 

of no use as a load carrying component. Therefore predicting the material characteristics of 

wood through NDT techniques is vital for the timber industry and has a long history of 

application in the wood products industry (Halabe et al, 1995). Moreover, the use of structural 

components is generally under standard applications which define the performance of the 

product in regard to its purpose. 

It has been shown that density, knots (size, frequency, and location) and modulus of elasticity 

(MOE) are the most suitable parameters for wood strength prediction. The modulus of 

elasticity has shown the best correlation to the strength for a single parameter (Steiger, 1996). 

Today a wide variety of NDT methods for wood are known and available on the market 

including: 

• radiography (X-ray and gamma-ray); 

• evaluation of visual characteristics; 

• near-infrared; 

• microwave; 

• ultrasonic; 

• acoustic; 

• machine stress rating (MSR); 

• reference testing (quasi-static test). 

Further technologies exist which are either not yet fully developed, or currently too complex 

or expensive to be used in commercial applications. 

Radiography 

In the 1980’s medical X-ray and gamma ray scanners were first used on wood. Scientists soon 

realised that the physical properties of wood are similar to the human body and thus 

developed a non-medical scanner which could be used in the timber industry. X-ray and 

5

gamma ray scanning identifies internal heterogeneities and defects in logs and sawn boards 

(Bucur, 2003). The rays are able to penetrate material without being reflected or broken. On 

the basis of this feature a precise map of the internal heterogeneities is created. Depending on 

the system it is possible to acquire a 2D or 3D image. Such industrial wood scanner systems 

require a high X-ray or gamma ray source. 

Gamma rays are of lower intensity than X-rays and can be used as a portable system for 

inspection of trees, poles and building elements (Bucur, 2003). The scan is done by irradiating 

the specimen from one or more sides and collecting the intensity of the transmitted radiation 

on the opposite side. The absorbed radiation is linked with the density and the moisture 

content of the material (Duff, 2005). Using this technology several strength affecting 

parameters like knots, density and moisture content can be determined contact free. The main 

advantages are that data are available in real time and a large volume of material can be 

rapidly inspected (Hanhijärvi et al, 2005). The disadvantage is that to ensure safety, the high 

intensity equipment has relatively large space requirements (Bucur, 2003). 

Evaluation of visual characteristics 

Visual inspection is one of the simplest and oldest methods of detecting exterior defects in 

wood members. The inspector observes the sample for parameters affecting strength such as 

knots, slope of grain and decay. This method requires good lighting, close attention from 

experienced operators and is limited to external degrade and features (Ross et al, 2006). 

Visual inspection is a relatively subjective method. 

Automatic optical scanning systems allow this process at production speeds. Normally such 

machines are equipped with different camera and laser systems. Such cameras are also used to 

identify the variation in surface colour along the boards (Duff, 2005). These systems both 

detect the existence of the defect and its position. The data is usually sent directly to a 

docking machine to optimise the cutting process. 

Near-infrared (NIR) methods 

The spectroscopy of near-infrared waves ranges from 800 to 2500 nm. The advantage is that 

NIR can penetrate deeper than mid-infrared radiation and is simple to operate (minimal 

preparation and rapid measurement). NIR has mainly been used for non-destructive testing of 

organic materials such as agricultural or food products. Today it is used mainly in the medical 

and chemical fields, but recently NDT methods using NIR have been tested in the timber 

industry. 

Tsuchikawa (2006) presented recent technical and scientific reports of NIR spectroscopy 

research in the wood and paper science industries. Others described a near-infrared method 

developed to predict the lignin content of solid wood. Strong correlations were found between 

the predicted lignin contents and the contents obtained from a traditional chemical method 

(Yeh et al, 2004). Schimleck et al (2002) described NIR measurements on small clear 

samples of Eucalyptus delegatensis and Pinus radiata to determine a number of physical 

properties including density, MOE, micro fibril angle and modulus of rupture (MOR). Good 

correlations were observed for all parameters, with R 2 ranging from 0.77 for MOR through 

0.90 for MOE to 0.93 for density. 

Microwave techniques 

Similar to X-ray and gamma ray, microwaves penetrate the wood and on the basis of wave 

reflection, and checks, splits and irregularities in the test specimen can be detected. Compared 

to X-ray and gamma ray, this technology is less expensive, but at current production speeds 

6

the NDT of timber by microwaves is impeded/disturbed by mechanical vibrations (Bucur, 

2003). Microwave testing is a contact free method that can detect defects and irregularity 

within the wood based on the determination of its dielectric properties which are principally 

due to the water content. Baradit et al (2006) described a microwave technique to generate 

and process data on knots in wood before processing. 

Stress wave methods 

The analysis of mechanical wave propagation in media of various complexities enables the 

measurement of elastic properties (Bucur, 2003). The use of acoustic methods, vibrational in 

the audible range (sonic) and at a frequency beyond human hearing capability (ultrasonic), for 

characterisation of the mechanical behavior of solid wood and wood-based composites is well 

documented. The velocity at which a stress wave travels in a member is dependent upon the 

properties of the member only. The terms sonic and ultrasonic refer only to the frequency of 

excitation used to introduce a wave into the member. All commercially available timing units, 

if calibrated and operated according to the manufacturer’s recommendations, yield to 

comparable results. 

The MOE and density are the main parameters which describe the wave propagation (Steiger, 

1996). The wave speed for isotropic and homogenous material at given physical conditions is 

defined by the following equation: 

Where, 

2 

2 

∂ u 1 ∂ u 

= 2 

2 

∂x 

V ∂t 

2 

x 

⇒ 

u is longitudinal displacement 

t is the time measured 

Ex is the modulus of elasticity 

ρ is the wood density 

Vx is the wave propagation speed 

7 

V 

x 

= 

Ex 

ρ 

(Equation 1) 

Basically there are two different methods to exploit stress wave velocity measurement 

(Krautkrämer, 1996): 

• Transmission technique where the ultrasonic wave is introduced by a transmission 

head into the test piece and received by a second receiver head. 

• The pulse echo method where a wave is reflected in a salient place (defect, surface) 

and the echo by the combined sending/receipt head is registered. 

In the wood industry the transmission technique is more commonly used to indicate the 

stiffness of a sample (Steiger, 1996). 

Ultrasonic 

Ultrasonic describes a stress wave with a frequency greater than the upper limit of human 

hearing, that is, high frequency (20 kHz – 600 kHz and higher) stress wave. This technique is 

particularly useful to detect decay as the stress wave propagation is sensitive to the presence 

of degradation in wood. In general ultrasonic stress waves travel faster in high quality and 

stiffness material than in material that is deteriorated or of low quality (Ross et al, 2006). 

Thus the wave propagation is based on physical and mechanical characteristics such as 

(Bucur, 2003): 

• MOE;

• density; 

• moisture content; 

• dimensions of the sample; 

• material type; 

• experimental conditions. 

Timber grading using ultrasound has been discussed by Sandoz (1989) and in Steiger, (1991). 

It was found that by measuring spruce (Picea spp.) beams of 10 cm by 22 cm cross section 

and 4.40 m length, a correlation of R 2 = 0.46 between wave velocity and MOR existed. 

Hanhijärvi et al (2005) reported R 2 s of 0.42 for strength and 0.57 for stiffness using a 

Sylvamatic strength grading machine. 

Sonic 

Measurement of acoustic velocity using the stress wave method is based on the same principle 

as the ultrasonic method. In the acoustic domain the input consists of a low frequency 

(approximately 1 Hz – 20 kHz) stress wave. 

The wave is introduced to the material by striking the specimen with an impact hammer. A 

force transducer connected to the specimen records the input signal. On the other side of the 

specimen an accelerometer is connected which receives an output signal. The measurement of 

these two signals (input and output), allows the stress wave velocity to be calculated. The 

analysis is based on a time analysis, which doesn’t require the knowledge of the boundary 

(Brancheriau, 2007). When striking the specimen a range of stress wave frequencies, and thus 

velocities, is inducted. As a consequence several velocities can be measured. 

Most of the algorithms used extract the fastest speed and the average speed of the group. On 

the basis of the velocity and sample density the dynamic MOE can be assessed. The wave 

velocity is expressed as: 

L 

Vx = (Equation 2) 

t 

Where, Vx is the wave propagation speed 

L is the distance between the two probes (sensors) 

t is the time-of-flight (TOF) 

The wave velocity is linked with the MOE and density and can be calculated by Equation 1 

displayed above. This expression is exact only for isotropic and homogenous materials at 

given physical conditions and is therefore only approximate for wood. Further parameters like 

energy loss can also be extracted in order to access the non-homogeneity of the tested 

material. A couple of commercial equipments mainly develop for standing trees stress wave 

velocity measurement are available (Fakkop FRS-06/00 or Director ST300 for example). 

Resonance method 

The most convenient method for measuring MOE with high precision depends upon 

measurements of the resonance frequencies in different modes (longitudinal, flexion or 

torsional) of simple structures for which the geometry and boundary conditions are known 

(Brancheriau and Bailleres, 2002). The fact that the technique is based on resonant structure 

ensures that frequency measurements will be precise. The technique can be extended to 

measure damping parameters and several signal descriptors. 

8

To illustrate these methods consider a prismatic, homogenous and isotropic beam with a 

length L, height h and width w. After an impact hits the beam either longitudinally or 

laterally, it vibrates freely in compression or bending respectively. For a longitudinal wave 

(also known as compression or P-wave) the particles vibrate parallel to the direction the wave 

is travelling. In transversal wave (also known as shear wave or S-wave), the motion of the 

particles is perpendicular to its direction of propagation. 

Because of these different kinds of movements, the longitudinal method is used to estimate 

the compression and tension characteristics, while the transversal method determines the 

bending characteristics. By measuring the movement of a vibrating beam the fundamental 

resonant frequency can be determined by a Fast Fourier Transform algorithm. The following 

expression shows the relationship between frequency and speed: 

n 

n = V , n∈ 

N* 

(Equation 3) 

2L 

f X 

Where, L is the length 

fn is the natural frequency (rank n) 

n is the frequency rank 

Vx is the wave propagation speed 

The dynamic modulus of elasticity along the longitudinal direction of the beam produced by a 

compression stress can be calculated using the following equation (Ibid): 

2 

2 fn 

E = 4L 

ρ 

(Equation 4) 

2 

n 

Where, E is the dynamic MOE 



n is the number of frequencies 

Using the transversal measurement, produced by a flexion stress, we can apply Bernoulli’s 

model which provides a solution to calculate the dynamic modulus of elasticity. This is 

achieved using the following equation (Ibid): 

E 

Gz 

2 

n 

4 

2 ρAL 

f 

= 4π 

(Equation 5) 

I P 

Where, E is the dynamic modulus of elasticity 


A is cross-section area 

L is the length 


Pn is solution of Bernoulli (rank n) 

IGz is the moment of inertia, which can be calculated for a rectangular section 

as follows: 

9 

n

3 

bh 

IGz = (Equation 6) 

12 

Where, b is base, horizontal dimension (length) 

h is height, vertical dimension (length) 

12 is the constant for moment of inertia of a member with rectangular cross 

section. 

In Timoshenko’s model for transversal / flexion vibration, the shear effect is no longer 

ignored. The solution according to Bordonné is more accurate for higher depth to length ratio 

beams. It allows the extraction of the shear MOE and uses more resonance frequencies 

information (Brancheriau and Bailleres, 2002). 

Recent studies (Ross et al, 2005; Tsehaye et al, 2000; Wang et al, 2003) have demonstrated 

that predicting MOE of trees, logs and sawn timber using acoustic velocities as well as 

resonance methods are highly correlated with the static MOE measured. Halabe et al (1995) 

found coefficients of determinations (R 2 ) between static bending MOE and stress wave MOE 

of 0.73 and 0.74 in the green and dry stage, respectively. 

Several commercial technologies are already available for acoustic grading on board such 

those from Microtec, Luxscan, Falcon engineering or Dynalyse AB for example. 

Remark on the measured acoustic velocities 

From above there are two different methods to measure the stress wave velocity: 

1- By measuring the time-of-flight (TOF) according to equation 2. The accuracy of TOF 

measurement depends on accurate identification of the arrival times of the acoustic 

wave signals, each from a start sensor (impact hammer) and a stop sensor 

(accelerometer). It depends on the quality of the signal recorded and the extraction 

algorithm used to detect the start and the stop signals. The MOE can be calculated by 

applying equation 1 provided that the density is known. The TOF method applied on 

standing tree is likely disturbed by dilatational or quasi-dilatational waves rather than 

one-dimensional plane waves. This leads to standing tree velocity being significantly 

higher than log velocities and skewed relationship between tree and log acoustic 

measurements. 

2- By measuring the resonance frequency according to equation 3. The resonance-based 

acoustic method is a well-established NDT technique for measuring long, slender 

wood members. The inherent accuracy and robustness of this method provide a 

significant advantage over TOF measurement in applications such as log 

measurement. In contrast to TOF measurement, the resonance approach stimulates 

many, possibly hundreds, of acoustic pulse reverberations in a log, resulting in a very 

accurate and repeatable velocity measurement. Because of this accuracy, the acoustic 

velocity of the logs obtained by the resonance-based acoustic method is usually a 

better predictor than the velocity obtained by measuring the TOF. The MOE can be 

calculated by applying equation 4 provided that the density is known. The constraint 

linked to this method is that the boundary conditions have to be perfectly known, this 

is not possible on standing tree. 

Machine stress rating 

Static bending is the foundation of MSR of timber. The development of such machines started 

in the 1960s and by 1963 the first industrial MSR machines were operating in the USA. These 

10

machines were characterised by a very high efficiency, but they could only sort planed timber 

with a maximum depth of approximately 40 mm. Therefore these machines found no practical 

application in Europe where larger dimensions are commonly produced and used (Krzosek, 

2003). 

MSR is currently the most common dynamic, mechanical load procedure. It measures the 

local stiffness of the material and sorts it into various MOE classes. The boards are fed 

through the machine flat wise, longitudinally and bent by rollers upwards and downwards in 

two sections. The span between the rollers is typically around 1.2 m. 

Depending on the design, the machine bends the boards to a constant deflection and measures 

the required force or the machine bends the boards with a constant force and measures the 

deflection. Using the load deflection relationship, the local MOE can be determined directly 

by using equations sourced from fundamental mechanics of material on every point of the 

board except for approximately the first and last 500 mm. This test method allows the 

stiffness profile of a board to be determined. Boards are usually graded using a combination 

of the lowest MOE (Lowpoint MOE), its position and the average MOE of the boards (Duff, 

2005). 

Static bending 

The resistance of a sample to slowly applied loads is measured by the static bending test. This 

procedure is generally conducted under standard conditions such as 20°C air temperature, 

65% relative air humidity and 12% wood moisture content. These values vary slightly 

depending on the standard used. The ends of the test sample are supported on rollers and a 

load is applied either centrally (three point bending) or two loads are applied in the middle 

third of the span (four point bending). In the past, three point bending tests were used to test 

small clear wood samples or wood composites, whereas four point bending tests were used to 

test full size specimens, although four point bending can also be used for small clear wood 

specimens. 

Because the values obtained by the different methods cannot be directly compared, 

Brancheriau et al (2002) presented an analytic formula using a crossing coefficient between 3 

point and 4 point bending. The deflection and load are measured at intervals using a 

deflectometer and load cell respectively. 

The first part of the curve is a straight line. There the deflection is directly proportional to the 

load and once the load is removed, the test specimen will return to its original state. This 

deformation is therefore in the elastic part of the beam (ε elastic). With increasing load a limit 

point of proportionality is reached (σp). Afterwards the deformation is no longer proportional 

to the load. With further load increase the material becomes plastic. This means when the load 

is removed, the beam will not return to its initial state (ε plastic). By increasing the stress to 

the point of maximum load (σu), the material begins to yield and fracture. The static MOE is 

determined from the slope α and the MOR value is equivalent to the maximum load attained. 

These values are considered as reference values. 

The test span of such measurements depends on the dimensions of the specimen. Normally it 

is 18 times the depth of the sample. Thus, long thin specimens need to be docked to the 

required length. In the different standards, a variety of methods for selecting the test specimen 

from a piece of timber is defined (Leicester, 2004). For example in EN 408 a selection from 

the low point of the board is required (EN408, 1995). 

11

Low point is the location within the board with the lowest grading modulus also known as 

biased position test sampling. Australian New Zealand Standard AS/NZS4063 specifies that 

the sample is selected randomly from a piece of timber (AS/NZS 4063, 1992). Random 

sampling and random position testing gives the direct comparison with the design values. 

However, the sampling error for strength data collected this way can be relatively high 

(Boughton, pers. comm.). 

Random position tests are useful to provide a good approximation to population average 

MOE. Biased position tests can focus on the low end of the distribution more effectively and 

are useful to determine the effectiveness of grading systems (Leicester, 2004). However, the 

data from biased position tests cannot be compared directly with the Australian design values 

so acceptance criteria for these tests must be found from comparative testing on each size and 

grade of product. These two methods may lead to considerably different results of mechanical 

properties. Leicester et al (1996) studied the equivalence between different in-grade testing 

methods. MOE and MOR were measured on about 150, 90 x 35 mm F5 and F8 grades of 

radiata pine boards with a length of 4.8 m. Small but significant differences of approximately 

20 percent at both the mean and the 5-percentile for strength were found. 

NDT of wood for grading: results from previous studies 

Brancheriau and Bailleres (2003) performed dynamic tests on 96 structural boards of larch 

(Larix europeaea). Through Partial Least Squares (PLS) analysis of acoustic vibrations in the 

audible frequency range, the researchers showed that the stiffness and strength could be 

accurately estimated. Further, they suggested that rapid, in-line grading systems could be 

developed relatively inexpensively through the installation of a vibration sensor, acquisition 

card and a computer for Fast Fourier Transforms and matrices’ calculations. 

One hundred southern pine (Pinus spp.) boards of dimension 100 x 50 mm and a length of 2.4 

m were tested in green and dry states by Halabe et al (1995). The velocity of longitudinal 

stress waves and the dynamic MOE were measured with transverse vibration equipment. 

Ultrasonic wave speeds were also measured and the static bending MOE was determined by 

using a four point bending machine. The samples were then dried to 12% moisture content 

and the described tests were repeated for dry specimens and at the end the failure strength was 

determined by four point bending equipment. The result of this research showed that the 

relationship between dry static bending MOE versus green stress wave velocity or the 

corresponding green MOE can directly be used to predict the dry static bending MOE (Ibid). 

In a study performed in New Zealand in 1998, 300 pine logs (species not reported but likely 

to be radiata pine) were tested using acoustics (Tsehaye et al, 2000). One end of each log was 

hit with a hammer containing an accelerometer that records the moment of impact. The sound 

wave passed along the 4.2 m long log and its arrival at the other end detected by a second 

accelerometer that was pressed against the log end. The 27-year-old pine logs from Mamaku 

Plateau in the Central North Island were sorted into three groups (27, 39, 27 logs), according 

to the speed of sound. After the measurements, the logs were sawn into 100 x 40 mm boards 

and then kiln dried to 12% moisture content. The MOEs of the dressed boards were 

determined by a stress-grading machine. The regression between the squared velocity of 

sound and the mean modulus of elasticity (300 logs) gave an R 2 of 0.46. Within the single 

groups the R 2 was between 0.45 and 0.57. 

From these results it follows, that acoustic sorting of logs provides the opportunity to send 

only the best quality, high stiffness logs, to the sawmill (Tsehaye et al, 2000). In this study 

only two parameters, speed of sound and MOE provided by a stress-grading machine were 

compared. The effective strength and stiffness of the boards was not determined. 

12

Ross et al (2005) conducted an investigation evaluating Douglas fir (Pseudotsuga menziesii) 

peeler cores. The longitudinal stress wave method was used to evaluate peeler cores from 111 

Douglas fir stems. Then the 2.6 m long peeler cores were sawn into 90 x 40 mm boards. The 

dynamic MOE of each board was determined in green, as well as dry condition (6-7% MC) 

by using stress wave and transversal vibration techniques. Afterwards, the boards were tested 

to failure strength in bending to determine the static bending MOE and the modulus of 

rupture. This study showed that stress-wave-predicted MOE of peeler cores is a good 

predictor of both dynamic and static MOE of timber obtained from the cores. Very strong 

relationships were found between stress wave and MOE of the peeler cores and dynamic 

MOE (stress wave MOE and vibration MOE) as well as static MOE (bending MOE and 

tensile MOE) of the timber. However, the correlations between stress wave MOE of the 

peeler cores and the bending and tensile strength (MOR) were low. This investigation has 

shown that a basic grading of specimens is possible using stress wave techniques (Ibid). 

Grabianowski et al (2006) performed a study about acoustic measurements on standing trees, 

logs and green timber of young Pinus radiata trees (aged 8 to 11-years-old). Thereby a time 

of flight tool (Fakopp ® 2D) and a resonance based system (WoodSpec) were used to evaluate 

the stiffness of the test material. The aim of this study was to determine how well acoustic 

measurements of green logs, estimate the properties of timber cut from those logs. The 

correlation using Fakopp ® 2D between the log values and those for the average of two boards 

from each log was 0.94. By using the resonance based system WoodSpec this relationship 

was 0.86. Significant correlations were found between the two tools, especially for stems (R 2 

= 0.96) (Grabianoski et al, 2005). 

During Combigrade Phase 1, Hanhijärvi et al reviewed the results from five previous 

investigations into NDT for strength prediction published between 1984 and 1997. Based on 

this review it was concluded that: 

• Correlations (coefficient of determination, R 2 ) varied between studies, probably due to 

differing materials and methods. 

• The highest correlation by any parameter tested achieved R 2 =0.7. 

• MOE is the best single variable for prediction of strength, followed by KAR and 

density. 

• A combination of predictors provides greater accuracy than a single predictor. 

Other relevant findings in the literature included: 

• Görlacher (1984) found that the natural frequency (dynamic MOE) correlated well 

with static test results, as did Blass and Gard (1994) in their tests on Douglas fir. 

• In separate studies Sandoz (1989) and Diebold et al (2000) found R 2 =0.45 and 0.53 

respectively for ultrasonic speed and strength. 

• On a small number of specimens Oja et al (2000) found a prediction of R 2 =0.41 for Xray 

(density and knot volume) and strength of sawn boards. 

• Similarly to the Combigrade project, Fonselius et al (1997) found that the accuracy of 

the predictors varies between species. For example in the Fonselius study it was found 

that knots explained 57% of the strength of pine compared to 27% for spruce. 

For Phase 1 of the Combigrade project (Hanhijärvi et al, 2005) approximately 100 logs each 

of spruce (Picea spp.) and pine (Pinus spp.) were investigated by testing them with different 

non-destructive testing methods. The logs were scanned by X-ray equipment and natural 

frequency and acoustic tomography measurements (only on 75 spruce logs, no pine logs), 

before they were sawn into nominal 150 x 50 mm boards. Approximately one board per log 

13

was selected and the batch was dried to an average of 10%. Final dimensions after drying and 

dressing were 146 x 46 mm, with a pencil round profile. 

A range of data was collected for the trial boards: 

• scanned for digital image analysis; 

• natural frequency measurements twice (two different operators); 

• X-ray scans; 

• acoustic-ultrasonic measurements; 

• sloping grain; 

• density by gross measure and by gamma ray; 

• Finnograder (gamma rays, microwaves, infrared radiation to predict strength); 

• Raute Timgrader MOE; 

• Compression MOE; 

• Average annual ring width by image analysis. 

The test material was loaded to failure in bending, and modulus of elasticity, bending 

strength, knot area and density were measured to determine grade. 

Findings from Phase 1 of the Combigrade project are summarised below: 

• Spruce and pine populations behave differently in regard to the predictors; 

• Stiffness parameters had the best single-variable predictions of bending strength: 

MOE measured by either static method, vibration method or by ultrasonic method. 

• Correlations between NDT density measurements (gross measurements, X-ray or 

gamma ray) and strength were within a similar range with X-ray providing the best 

value. 

• Density is a better grade predictor for pine than for spruce. 

• Knot parameters provide good predictions of strength and density for pine, but not 

spruce 

• Irradiation equipment (X-ray and gamma ray) provide slightly better strength 

prediction than surface inspection such as KAR. 

• Sloping grain measurements didn’t have the potential to predict strength. 

• Relatively strong correlations were obtained from log measurements with destructive 

board tests. 

• Log X-ray and dynamic MOE based on natural frequency both provide R 2 of 0.60 or 

better for pine. 

• Combinations of devices provided correlations of R 2 =0.80 to 0.85 for pine and 0.60 to 

0.65 for spruce. 

• Combination of knot measurements with density and annual ring width provides 

effective predictions for strength with R 2 =0.7 (pine) and 0.6 (spruce). 

Based on recommendations from Phase 1 of the Combigrade project, a larger sample with 

replicates of different sectional sizes was tested to form the Phase 2 follow up trial 

(Hanhijärvi et al, 2008). For Phase 2 more than 1000 logs each of spruce and pine were 

sampled and after NDT data gathering from the logs, one board per log was selected from a 

mix of seven different size classes. In addition to the larger sample and range of dimensions, 

Phase 2 boards weren’t planed after kiln drying. Results determined during Phase 2 of the 

Combigrade project can be summarized as: 

• The two species behaved differently, similar to the small sample tested during Phase 

1, with stronger correlation between predictors and strength usually achieved by pine. 

14

• Cross-section dimensions also behaved differently, providing an insight to how size 

affects correlations. 

• The correlation of density to MOE and MOR in bending decreases with an increase in 

section size. 

• The results and conclusions for correlations’ analyses were similar to the results from 

the smaller sample tested in Phase 1. 

Standard requirements 

The Australian Standard AS1720.1:1997 Timber structures, Part 1: Timber properties defines 

structural performance of machine graded pine (MGP) by the 5th percentile strength (MOR) 

of a sample population (random position tested) and by an average MOE (Standards 

Australia, 1997). In the current AS/NZS4063, the design properties are related to 

characteristic strength and MOE: 

• characteristic strength is lower 75% Confidence limit of the 5 th percentile strength of 

the population estimated from tests on a sample; 

• characteristic MOE is 75% Confidence limit of the average population MOE 

estimated from tests on a sample. 

To achieve these requirements, MGP material needs to be graded on the basis of 

AS/NZS1748:2003 Timber-Stress-graded-Product requirements for mechanically stressgraded 

timber. This Standard requires that every board is initially graded by a mechanical 

stress-grader which sorts the timber on the basis of its MOE (low point or mean or 

combination) and secondly by a visual inspection to detect strength-limiting and utilitylimiting 

characteristics. 

The machine stress-grader settings are determined by continuous monitoring (based on 

random position testing reference) of the values obtained from batches. Strength-limiting 

parameters are knots, resin- and bark pockets, cross- and heart shakes and splits. Utilitylimiting 

parameters for softwood species include dimensional tolerance, squareness, knots, 

wane and want, machine skip, and distortion (bow/spring/twist). In the visual over-ride, high 

attention must be paid to the board ends, where the machine stress grader cannot determine 

the MOE (Standards Australia, 2003). 

The European standard EN14081-1:2003 Timber structures – Strength graded structural 

timber with rectangular cross section – Part 1: General requirements requires either visual or 

machine graded timber. Visually graded timber accounts for: 

• strength-reducing characteristics like knots, slope of grain, density, rate of growth and 

fissures; 

• geometric characteristics like wane and warp; and 

• biological characteristics like fungal and insect damage. 

If the timber is machine graded the boards have to be graded on their full length. If that is not 

the case, as in bending type machines, the non-graded part needs to be visually examined (EN 

14081-1, 2003) as for the Australian Standard. 

The European standard requires a biased position test as the reference static bending test. 

15

Study Methodology 

Materials 

Sampling method 

Logs provided for the trials were selected from a mix of butt logs (30%) and upper logs (70%) 

in order to over-sample the low grades and ensure sufficient representation of the grades 

below MGP10 and in the lower half of the MGP10 population. Rather than sample 1 

board/log and have a large number of individual logs, this project aimed to maximize the 

number of boards recovered from each log to provide the best experiment for comparing 

results from the NDT log tests and the subsequent results from board testing. 

Paper log end templates based on Smith et al (2003) were adhered to both ends of each log to 

enable identification from log (Figure 1) through the green and dry chains and subsequent 

testing. Information provided on each label enabled the unique log number to remain on sawn 

boards through all stages of processing and testing as well as determining the log end (small 

end or large end) and relative in-log position of each board. 

Figure 1. Log end template. 

Stage 1 Stiffness limited radiata pine (Radiata E) 

Stiffness limited Pinus radiata, radiata pine was sourced from 67 logs from two plantations 

located near the Victoria-New South Wales border and aged approximately 28-years-old. The 

average centre diameter for the batch was 370 mm (range 280 mm to 500 mm). From this 

batch the target number of board samples was 600. The sawmilling process provided 992 

green boards, of which complete data sets for all trials were compiled for 517 boards. An 

overview of the testing conducted on this batch is provided in Figure 2. 

16

stage 1 

Carter Holt Harvey Myrtleford 

DPI&F 

Brisbane 

DPI&F 

Brisbane 

67 radiata pine logs 

Three vibration measurements 

Assortment of the boards 

52 x 100 mm 266 pieces 

42 x 100 mm 548 pieces 

42 x 80 mm 9 pieces 

Sawing the logs and numbering the boards 


25 x 100 mm 153 pieces 

25 x 80 mm 9 piecess 

Bing® vibration measurements on green boards 

Determination of MOE and a range of signal descriptors 

Docked all the boards to a maximum length of 5.0 meters 

and resaw the 52 mm boards to 42 mm to simplify further 

tests 

Random sub-sample of 600 boards (42 x 100 mm) 

Weyerhaeuser Caboolture 

Hyne Timber 

Melawondi 

DPI&F Brisbane 

17 

Dimensions and density measurements by gamma ray 

MOE by Thumper in-line vibration system 

Drying and dressing the boards (? 12% moisture content / 35 mm by 90 mm) 

Metriguard (MSR) testing at Weyerhaeuser 

Determination of MOE along the boards 

WoodEye scanner 

Determination of external defects 

Bing® vibration measurements on dry board 


Four point reference bending test at Salisbury (Biased & Random) 

Determination of MOE & MOR 

Figure 2. Sample flow for Stage 1, stiffness-limited radiata pine. 

Stage 2 Caribbean pine (Caribbean) 

One-hundred and sixteen logs representing the sub-tropical exotic pine resource of south-east 

Queensland were provided for the trial. The age of the source plantation was not provided by 

the supplier who estimated that the trees were 25- to 30-years old. The average log diameter 

for this batch was 250 mm (range 230 mm to 270 mm). The target number of boards from this 

sample was 400 and from 741 green boards, 489 were used to provide the data for the batch. 

An overview of the testing is provided in Figure 3 below. 

Stage 2 

QPIF Brisbane, Weyerhaeuser Caboolture 

QPIF 

Brisbane 

106 Caribbean pine logs 

Three vibration measurements Drying and dressing the boards (? 12% moisture content / 35 mm by 90 mm) 




100 x 40 mm 595 pieces 




Figure 3. Sample flow for Stage 2, Caribbean pine. 

Weyerhaeuser Caboolture 

QPIF 

Brisbane 

Hyne 

Timber 

Melawondi 

QPIF 

Brisbane 

Metriguard (MSR) testing at Weyerhaeuser 




WoodEye® scanner 




Stage 3 Strength limited radiata pine (Radiata R) 

Forty-seven logs with an average centre diameter of 420 mm (range 300 mm to 580 mm) 

representing the strength-limited radiata pine resource of Western Australia were provided for 

testing with a target sample of 400 dried boards. After sawing and drying, 582 boards 

provided the data set for this resource sample. The trial flow is depicted in Figure 4.

Stage 3 

Wespine Bunbury 

Wespine 

Bunbury 

47 radiata pine logs 

Three vibration measurements Drying and dressing the boards (? 12% moisture content / 35 mm by 90 mm) 


100 x 40 mm 642 boards 




Wespine Bunbury 

DPI&F 

Brisbane 

Hyne 

Timber 

Melawondi 

DPI&F 

Brisbane 

18 

Metriguard (MSR) testing at Wespine 

Grain angle, moisture content, and unsound wood position measurements by X-Ray 




WoodEye scanner 


Figure 4. Overview of testing, Stage 3, strength-limited radiata pine. 



Kiln drying 

After completion of tests in the green (unseasoned) condition, boards were transported to a 

commercial softwood plant for high temperature kiln drying to a target average moisture 

content of 12%. After equalisation, the test boards were dressed to 90 x 35 mm. 

Equipment and methods 

Non destructive testing- mechanical properties measurement 

Machine stress rating 

Metriguard Continuous Lumber Tester (CLT 7100) equipment was used to collate MOE 

profiles for Stage 1 (radiata E) and Stage 2 (Caribbean) dry boards and Metriguard High 

Speed Continuous Lumber Tester (HCLT) for the Stage 3 (radiata R) dry boards in order to 

prescribe stress ratings. The boards were bent flatwise by rollers downward and then upward 

as depicted in Figure 5 below. 

Figure 5. Schematic diagram of Metriguard CLT (source: Web 4) 

Thereby the bending force and the deflection in both bending sections were measured. The 

local MOEs at intervals of 13.9 mm were automatically calculated, from which the average 

and the low point MOE were provided on the full length (excluding the leading and trailing 

820 mm ends sections of boards). Additionally, the Metriguard MOE profile for the static test 

section was extracted to allow correlation with the static bending MOE and MOR test results.

The predictors used for the analysis were: 

• full board profile extraction: Metri_avg board and Metri_min board; 

• static bending span: Metri_avg Static Test and Metri_min Static Test. 

In-line acoustic test 

The Thumper is an in-line measurement device used to capture the longitudinal frequency and 

combine the data with parameters captured by the gamma-ray and laser measuring tools in the 

same green chain to calculate the dynamic MOE of each green board. It consists of a 

mechanical pneumatic cylinder that strikes the board end and a microphone which receives 

the signal. In front of the cylinder is a laser proximity switch which senses the presence of the 

board and triggers the firing of the cylinder to strike the board. The impact sets up a stress 

wave which travels immediately through the board and is detected by the microphone. With 

the vibration fundamental frequency collected by the microphone and the dimensions as well 

as the density provided by the gamma ray system, the dynamic MOE of each board can be 

calculated, allowing sorting of the green boards into stiffness classes. 

The in-line Thumper system delivers a vibration MOE parameter (Thumper_MOE). 

Off-line acoustic test 

The acoustic velocity and vibration measurements were captured using Bing ® and WISIS 

(Wood in Situ Inspection) products which were developed by CIRAD (http://www.xylometry.org/en/softwares.html). 

WISIS measures the time of flight from an induced stress wave, based on a time-frequency 

analysis, not on a frequency analysis like the Bing ® procedure. From the extracted results the 

MOE can be calculated by using Equation 1. This type of measurement can also be applied to 

standing trees to predict wood stiffness similar to commercial tools currently available in the 

forest wood quality sector. With velocity measurement, no boundary conditions are required, 

contrary to vibration analysis. 

Bing ® allows determination of the bending and compression MOE by analysis of the natural 

vibration spectrum of a piece of wood. The technique is also known as the resonance method 

as it allows determination of resonance frequencies of a beam from its response to an impact. 

The two systems consist of the devices shown below in Figure 6 and associated software. 

19

F 

G 

C 

D 

Figure 6. Equipment used for the acquisition of the off-line acoustic signals. 

A Laptop or Desktop 

B Data acquisition card (Pico Technology Type Picoscope 3224) 

C Accelerometer (Brüel & Kjaer Type 4397) 

D Impact hammer (Brüel & Kjaer Type 8206-2) 

E Conditioner (Endevco Type 4416B) for accelerometer and impact hammer 

F Low pass filter to avoid aliasing effect 

G Screw for the accelerometer attachment (magnet system) 

Logs and green boards were tested using the Bing ® for: 

• longitudinal vibration- 

o compression waves, 

o frequency 2000 Hz, 

o four resonance frequencies viz F1 to F4, and 

• transverse vibration- 

o flexion waves, 

o frequency 1000 Hz, 

o five resonance frequencies viz F1 to F5. 

A 

B 

These measurements provided the following range of vibration signal descriptors: the MOEs 

in each configuration were calculated according to the equations described in the paragraph 

“Resonance method” above. 

The vibration spectra were analysed in order to provide the following range of vibration 

signal descriptors: 

• dynamic MOE associated with Fn (MOE_n). 

• spectral centre of gravity divided by the fundamental frequency, F1 in % 

(SCGravity). 

• spectral extent of gravity divided by the fundamental frequency, F1 in % 

(SBandwidth). 

• spectral slope divided the fundamental frequency, F1 in % (SSlope). 

• quality factor (inverse internal damping or friction) associated with Fn (FacQn). 

20 

E 

E

• inharmonicity factor (IF). 

• mean power in frequency sub-band; between 0 and f1; f1 and f2... (MSPower). 

• sub-band energy ratio, between the resonance frequencies, i.e. sub-band energy/total 

range energy (MSNrjRatioF_n). 

• mean power of the sub-band, resonance shoulder defined by a pass band of 20dB 

(Pow_n). 

• sub-band energy ratio, resonance shoulder defined by a pass band of -20dB 

(MSNrjRatioF_n). 

• MOE extraction using Timoshenko’s model and Bordonné’s solution (MOET). 

A full description of the parameters is provided in Appendix 1. 

Maximal relative error on vibration MOEs 

After the signal conversion from analogue to digital, the Eigen frequencies were calculated by 

means of the Fast Fourier Transform. The resolution (r) is a function of sampling frequency 

(Fe) and of the number of measurement points (N): 

f e r = (Equation 7) 

N 

The absolute error Δf can be increased by the experimental absolute error (1 Hz) and half of 

the resolution. Therefore the maximal relative error for Bing ® measurements can be 

determined as follows: 

Compression: 

Flexion: 

Δf Δf 

i i _ exp + 0. 

5 ⋅r 

= 

f f 

i 

Resolution r = 1.19 Hz 

Δf i _ exp = 1. 

0 Hz 

i 

(Equation 8) 

Δf 

1 

f1 = 193 Hz = 0. 

83% 

f 

Δf 

2 

f2 = 583 Hz = 0. 

27 % 

f 

Δf 

3 

f3 = 869 Hz = 0. 

18 % 

f 

Δf 

4 

f4 = 1161 Hz = 0. 

14 % 

f 

Resolution r = 0.3 Hz 

Δf i _ exp = 1. 

0 Hz 

1 

2 

3 

4 

21

Δf 

1 

f1 = 10.4 Hz = 11. 

1% 

f 

Δf 

2 

f2 = 22.6 Hz = 5. 

1% 

f 

Δf 

3 

f3 = 54.2 Hz = 2. 

1% 

f 

Δf 

4 

f4 = 92.1 Hz = 1. 

2 % 

f 

1 

2 

3 

4 

The maximal relative error of the dynamic MOE can be calculated by the following equation: 

ΔMOE 

MOE 

Bing , L 

Bing , L 

Δm 

Δh 

Δb 

ΔL 

Δf 

1 

= + + + + 2 

m h b L f 

22 

1 

(Equation 9) 

For the mass: m = 5.4 ± 0.02 kg For the width: b = 30 ± 1 mm 

For the height: h = 90 ± 1 mm For the length: L = 5000 ± 20 mm 

Thus the maximal relative errors on compression measurements are: 





Bing , L , 1 

Bing , L , 1 

Bing , L , 3 

Bing , L , 3 

= 

= 

6. 

4 % 

5. 

1% 

ΔMOEBing 

, L , 2 

, = 5. 

3% 

, 


Bing , L , 2 


, L , 4 

, = 5% 


Bing , L , 4 

For MOEBing, F, 1-4 the maximal relative errors are: 





Bing , F , 1 

Bing , F , 1 

Bing , F , 3 

Bing , F , 3 

= 

27% 


, F , 2 

, = 15% 

, 


Bing , F , 2 


, F , 4 

= 9% 

, = 7% 


Bing , F , 4 

The data processing of the acquired digital signal involved the use of zero padding and 

smoothing procedures, to provide a more accurate reading of the resonance frequency. 

Gamma ray 

An in-line gamma-ray system (Geological & Nuclear sciences Ltd) was used to capture 

estimated density and moisture content data. The boards move transversely and are pushed 

along by lugs at 450 mm intervals on a chain conveyer system. The speed of the system can 

be varied up to 100 boards/minute, but is typically around 70-80 boards/minute. A system of 

four laser range instruments, positioned above and below the line, is used to measure the 

thickness, width and the length of the piece of timber. This information is combined with the 

gamma ray result to provide the average green density of the board. Gamma ray 

measurements are made simultaneously at four positions along the board by using four 

separate heads positioned between the chains. These measurements are combined to give

average values which are used to sort the boards for subsequent kiln drying or machine stress 

grading. 

Non destructive testing- structural properties measurement 

Linear high grader (LHG) 

The LHG was developed by Coe Newnes/McGehee ULC (Canada). Coe Newnes/McGehee 

ULC provides solutions for sawmill systems, planer mill systems, scanning and optimization 

technology (Web 5). 

The functions of the LHG are summarised as: 

• Measures wane, skip and holes. 

• Measures grain angle, moisture content, and unsound wood locations. 

• Determines best grade/length based on mill’s grading rules. 

• Marks board after scanner – ID number. 

• Read boards ID. 

• Sends trim and grade decision. 

The LHG utilises X-ray technology to analyse density variation allowing the identification of 

knots within boards. Data are combined with low-point MOE results determined by 

Metriguard HCLT equipment to predict MOR. The LHG scanning was only used on the 

Radiata R resource. 

LHG KAR is the LHG’s estimation of knot area ratio (KAR). A second parameter was 

extracted from the knot position data obtained with the X-ray, called LHG I ratio which is 

the ratio of the inertia (I value) of the knots in the window to the inertia of the full cross 

section. The maximum value of I ratio was extracted using a Gaussian sliding-window. 

LHG Weighted KAR is the same as LHG KAR, but only for knots in the outer quartile of 

the depth of the beam, where the outer quartile has weighting of 1.0 and the inner quartiles 

have weighting of 0.1, again using a Gaussian sliding-window. 

LHG prediction of strength (LHG predicted Strength) was derived from the combination of 

LHG parameters and Metriguard HCLT profile from the biased and random test span. All 

these parameters were extracted locally from the data gathered from the static test span. 

WoodEye ® 

A WoodEye ® optical scanning system was used to record defect type, size and location 

features. The WoodEye ® scanning system used combines: 

• grey scale camera; 

• RGB colour camera; 

• tracheid effect LASER; and 

• profile detection LASER. 

The sensor-system scans each piece on all four sides for natural features such as black knot, 

sound knot, fibre knot as well as board geometry. Other features such as wane and pith 

weren’t considered during this study. 

For strength grading the knot categories are the most interesting. Black knot (Bk) and Sound 

knot (Sk) refers to knot defects created from the grey scale sensor. Fibre knot (Fk) is created 

from the laser sensor and is based on detection of a fibre disturbance. Therefore, a certain 

23

knot can give rise to both a Fibre knot and a Black/Sound knot. The sizes can differ due to the 

detection situation. Fibre knots tend to be larger. The contour of the combined knot silhouette 

(WE) was used as a predictor. 

The beam profile construction is based on a rendering discrete process over the length of the 

beam with 1 cm steps. Each profile step is associated with a defects’ projection of a portion on 

the length of the beam equivalent to two times the height of the beam. The projection window 

can be: 

• rectangular on the entire beam portion; 

• rectangular external: 1/3 of the top and bottom part of the beam height; 

• rectangular interior: 1/3 of the centre part of the beam height; 

• triangular according to a diamond-shaped pattern. 

All the profiles are filtered with a Blackman sliding-window of which the size is equivalent to 

the beam height. Descriptions of the WoodEye ® profiles are provided in Appendix 2. 

Destructive standard static bending tests 

Static bending tests were performed using a testing method in accordance with AS/NZS 

4063:1992. Biased position tests were used on every board, and where the random position 

test location was within the useful remnant of the board after biased testing, a random position 

test was also performed on the same board. MOE and MOR were calculated for each test. 

Protocols used for the selection of biased and random samples are described in Appendix 3. 

The load for the reference testing was applied and measured with a Shimadzu UDH-30 tonne 

(300 kN) universal testing machine depicted below (Figure 7). 

Figure 7. Shimadzu UDH-30 t Universal testing machine. 

The term ‘universal’ indicates that the machine is capable of performing a range of tests 

including static bending, tension, compression, shear and hardness tests on large samples. The 

support consists of a solid steel roller 240 mm long by 50 mm diameter and a flat mounting 

plate. The mounting plates have two holes which are used to locate the plate over machine 

24

olts situated in the centre of the supports. The roller, and to a lesser extent the mounting 

plate, are held in place by two springs which attach to bolts protruding from the centre of the 

load roller. The Shimadzu UDH-30 is a ‘Grade A’ testing machine in accordance with 

Australian Standard AS 2193:2002 Calibration and classification of force-measuring systems. 

The Australian and New Zealand Standard, AS/NZS 4063:1992 Timber-Stress-graded-Ingrade 

strength and stiffness evaluation, requires that the specimens shall be conditioned to a 

temperature of 20 ±3°C and in an environment having a relative humidity of 65 ±5%. This 

conditioning shall continue until the moisture content is stable within each piece (10-15%). 

Moisture content was checked using an electric resistance moisture meter to confirm that the 

boards had conditioned to the range specified in the standard. Bending strength and stiffness 

were then tested according to the methods specified in AS/NZS 4063:1992. 

In the middle of the span the deflection was measured with a strain gauge type linear 

displacement transducer. The bending test span was 1620 mm with load applied at four points 

and the span-to-depth ratio was 18:1. The load deflection curve was measured up to 1.6 kN 

for all specimens. The modulus of elasticity can be determined from the slope of the linear 

relationship between the applied load (P) and the resulting deflection (ε) using the following 

equation: 

3 

23 l s ΔP 

MOE = ⋅ ⋅ 3 

108 bh ε 

25 

(Equation 10) 

After removal of the displacement transducer, loading continued until failure of the specimen. 

The modulus of rupture was calculated using the following equation: 

l P 

bh 

s MOR = 2 

(Equation 11) 

The following equation gives an estimation of the relative maximal error on static bending 

MOE: 



stat 

stat 

Δb 

Δh 

Δl 

s Δk 

= + 3 + 3 + (Equation 12) 

b h l k 

The relative errors on the measurements are based on a sample of 60 specimens and are: 

b = h = ± 1mm 

= ± 10 mm 

l s 

k is calculated using the 95% confidence interval from the slope of the force-deflection 

diagram. This method ensures the maximum relative error in the calculation of static MOE 

using Equation 12. 

Δ k 

The coefficient is calculated as 3 % and due to the similar methodology used in this 

k 

study the same value was used. Thus the maximal relative error for static MOE is: 



stat 

stat 

≈11% 

The maximal relative error of the MOR can be calculated by the following equation: 

s

ΔMOR 

Δb 

Δh 

Δl 

s ΔP 

= + 2 + + 

MOR b h l s P 

The relative error of the load applied is known as 1 % for the machine used. Therefore the 

maximal relative error of the modulus of rupture is: 

ΔMOR 

MOR 

≈ 6% 

Biased versus random position testing 

A biased test involves selection of a particular feature or position on the piece and deliberate 

placement in the centre of the test span in order to focus the results on the feature selected. 

For edge-biased bending tests, the selected feature is placed on the tension edge. 

A random test is where the central position of the test span is allocated using a random 

number generator to determine the measurement datum along the board without consideration 

of any properties or features of the timber in allocating test position. 

The issues of grade effectiveness must consider two opposing facts: 

• AS/NZS4063 is currently being revised, but will be underpinned by evaluation of 

characteristic values based on random-position testing. As a result, verification testing 

of stress-graded products will continue to be based on random-position tests. 

• The principle of stress-grading is founded on detecting the weakness in each piece to 

avoid failures in service where every part of most pieces will be used in some sort of 

role in buildings. The best grading methods will identify the lowest local strength of 

each piece rather than its global strength based on a random positioned test. 

The implications for grade evaluation are: 

• In performing normal commercial grading, the verification testing will normally use 

random-position tests and the test results will feed back to the threshold settings to 

ensure that the design characteristic values are reliably obtained by each graded 

product. 

• In comparing grading methods, it is the correlation of grading parameter and the 

biased strength that is important. The most effective methods have higher correlation 

coefficients as they are most able to correctly predict the lowest local strength of each 

piece. 

For this study, most use will be made of the biased position test data. Successful grading 

methods will have a high correlation with the biased strength. Although it is possible to 

achieve meaningful results from random-position tests, very large sample sizes are required. 

The scatter of the results is generally high, as the test span may not contain any of the limiting 

material that actually determined the grade of the piece. 

Simulated MGP grades recovery by best predictors 

The static bending values obtained from the biased samples were used to find the best 

correlation for each non-destructive measurement system with the collected grading modulus. 

These correlations were identified by simple or multiple linear regressions. 

The grading modulus which provided the best correlations to the static bending data, were 

then used to allocate the boards into MGP grades. Thresholds were adjusted until the average 

of the MOE random and the 5 th percentile were equal or above the standard requirements. The 

average and 5 th percentile were calculated on the basis of the MOE and MOR obtained from 

26

the random sample tests. The outputs of various grading measures were used to group the test 

timber into notional grades and the test results used to confirm that the properties of each 

notional grade exceeded the design values. 

If the thresholds were adjusted for each group, one would optimise the recovery and the grade 

limiting properties would be much the same for each group, but the recovery would be 

different between the groups. This method would be very hard to implement in practice as a 

producer would have to know the optimum threshold to use before starting production, so that 

all of the pieces could be graded to give the grade limiting property just higher than the design 

value. 

A range of different thresholds was used and all of the results for each resource and for each 

prediction were plotted. 

Data extraction and statistical approach 

Automated data extraction was undertaken through collaboration with CIRAD Xylometry, 

Montpellier France. Mathematical and chemometrics’ statistical tools were used to provide 

correlations between data sets and to develop efficient strength and stiffness prediction 

equations. These algorithms were derived from raw and combined parameters using multivariate 

linear regression (MLR) and partial least squares (PLS). 

The following discussion on MLR and PLS was modified from statsoft.com (Web 2). 

For many data analysis problems, estimates of the linear relationships between variables are 

adequate to describe the observed data, and to make reasonable predictions for new 

observations. When the factors are few in number, are not significantly redundant (collinear), 

and have a well-understood relationship to the responses, then MLR can be a good way to 

turn data into information. 

The MLR model serves as the basis for a number of multivariate methods such as: 

• discriminant analysis, i.e. the prediction of group membership from the levels of 

continuous predictor variables; 

• principal components regression, i.e. the prediction of responses on the dependent 

variables from factors underlying the levels of the predictor variables; 

• canonical correlation, i.e. the prediction of factors underlying responses on the 

dependent variables from factors underlying the levels of the predictor variables. 

These multivariate methods all have two important properties in common in that they impose 

restrictions such that: 

• factors underlying the Y and X variables are extracted from the Y'Y and X'X matrices, 

respectively, and never from cross-product matrices involving both the Y and X 

variables, and 

• the number of prediction functions can never exceed the minimum of the number of Y 

variables and X variables. 

PLS has become a standard tool in chemometrics for modelling linear relations between 

multivariate measurements and is particularly suitable for monitoring and controlling 

industrial scenarios which may have hundreds of controllable variables and dozens of outputs. 

PLS represents a compromise between principal component regression (PCR) and MLR. PCR 

detects factors that account for the greatest amount of variance in the predictor variables. 

MLR seeks uncorrelated variables that best estimate the predicted variable. PLS screens for 

factors which account for the variance and achieve the best correlation. A great advantage of 

27

PLS over other regression methods is that it can handle large numbers of correlated variables 

as predictors. The method is based on projections whereby a set of correlated variables is 

compressed into a smaller set of uncorrelated variables. 

Partial least squares is a method for constructing predictive models when the factors are many 

and highly collinear. The emphasis is on predicting the responses and not necessarily on 

trying to understand the underlying relationship between the variables. When prediction is the 

goal and there is no practical need to limit the number of measured factors, PLS can be a 

useful tool. Partial least squares regression is an extension of the multiple linear regression 

model (e.g., Multiple Regression or General Stepwise Regression). In its simplest form, a 

linear model specifies the (linear) relationship between a dependent (response) variable Y, 

and a set of predictor variables, the X's, so that 

Y = b0 + b1X1 + b2X2 + ... + bpXp (Equation 13) 

In this equation b0 is the regression coefficient for the intercept and the bi values are the 

regression coefficients (for variables 1 through p) computed from the data. 

Partial least squares regression is probably the least restrictive of the various multivariate 

extensions of MLR. This flexibility allows it to be used in situations where the use of 

traditional multivariate methods is severely limited, such as when there are fewer 

observations than predictor variables. Furthermore, PLS regression can be used as an 

exploratory analysis tool to select suitable predictor variables and to identify outliers before 

classical linear regression. 

An important factor to consider when interpreting the results provided in this report is that the 

variation of the coefficient of determination is expressed as an absolute difference on a scale 

of 0.00 to 1.00, as opposed to a relative comparison. 

28

Results 

Static bending 

Biased and random tests together 

Table 1 displays basic statistics for dry density, MOE and MOR at 12% moisture content 

(MC) for all the boards tested (biased and random together) for the three resources. Caribbean 

has the highest density followed by Radiata R with Radiata E having the lowest density. The 

MOE follows logically the same trends, with a 5% difference between Radiata E and Radiata 

R and 21% between Caribbean and Radiata E. Caribbean MOR is 34% higher than Radiata R. 

The MOR for Radiata R is 3% lower than Radiata E MOR despite its higher MOE. The low 

Radiata R MOR confirms the resource quality observation by the West Australian industry 

which considers this resource as strength limited. 

The standard deviation of Caribbean density is significantly higher than the other two 

resources. MOE and MOR standard deviations of Radiata R and Caribbean are very close 

whereas Radiata E shows comparatively low standard deviations for both properties. 

Table 1. Descriptive statistics of density and mechanical properties for the three resources, biased and 

random combined. 

Resource Mechanical properties Mean Std. Deviation N 

Radiata E Dry density (kg/m 3 ) 486 51 724 

MOE (MPa) 8317 2450 723 

MOR (MPa) 33 16 724 

Radiata R Dry density (kg/m 3 ) 508 41 689 



Caribbean Dry density (kg/m 3 ) 563 71 896 



The bivariate Pearson's correlation coefficients between density, MOE and MOR are 

displayed in Table 2 below. 

The MOE vs density correlations are quite different for each resource. Radiata E has the 

highest correlation coefficient followed by Radiata R with Caribbean having the lowest 

correlation coefficient. A similar trend is observed for MOE vs MOR. 

29

Table 2. Bivariate Pearson's correlation coefficients between density and mechanical properties for the 

three resources, biased and random combined. 

Resource Mechanical properties Dry density (kg/m 3 ) MOE (MPa) 

Radiata E MOE (MPa) 0.68 

MOR (MPa) 0.54 0.80 

Radiata R MOE (MPa) 0.54 


Caribbean MOE (MPa) 0.31 


Note: All the correlations are significant at the 0.01 level (2-tailed). 

Figure 8 shows the linear correlations between MOE and density for all the resources. From 

the scatter plot it appears that the Caribbean resource displays a significant number of outliers 

which have a high density with proportionally a low MOE. These outliers boards can be 

explained by the presence of a large quantity of resin and the occurrence of compression 

wood (high MFA) which both increase the density without significantly increasing the 

mechanical properties. Radiata E and Radiata R resources are different: the data shapes are 

not similar with Radiata E having a higher average MOE in the low density range. 

Figure 8. Linear regression between MOE and density for the three resources. 

The density vs MOR correlations show the same trend but exacerbated (Figure 9). 

30

Figure 9. Linear regression between MOR and density for the three resources. 

The MOR vs MOE linear correlations for all the resources are displayed in Figure 10. The 

general observation is the “trumpet” like shape of the data scatter points, in other words 

residual error on MOR increases with the MOE. This could be a source of heteroscedasticity 

problems (non-homogeneous variances) when developing linear regression equations. 

As expected the boards with the lowest strength belong to Radiata R resource. They are 

spread on a quite large span of MOE, up to approximately 12000 MPa. This induces a sort of 

“belly” on the Caribbean resource scatter plot. This is characteristic of a strength limited 

resource. 

The Caribbean resource has the lowest coefficient of determination which can be explained by 

a large variation of MOR values for a given upper range MOE (above 14000 MPa). Again the 

presence of a large quantity of resin associated with resin checks and the occurrence of 

compression wood could explain this observation. 

31

Figure 10. Linear regression between MOE and MOR for the three resources. 

Random and biased 

Table 3 displays the three resources’ dry densities, MOEs and MORs at 12% MC for all the 

boards tested by random and biased tests separately. 

The number of biased boards is larger than for random boards since the priority was given to 

biased position when performing the static bending test. The protocol was slightly different 

for Radiata E resource which explains the lower recovery of random boards (see Appendix 3), 

40% comparing to an average of 60% for the two other resources. 

The general tendencies observed when comparing the resources on all the boards taking all 

test positions together are similar to those observed on random or biased test positions. The 

mean density between biased and random is not significantly different whereas random and 

biased MOE and MOR are obviously quite different. 

The total average MOE variations from biased to random relevant to the average MOE of all 

boards are 17%, 22% and 5% for Radiata E, Radiata R and Caribbean respectively. The total 

average MOR variations from biased to random relevant to the average MOR of all boards are 

54%, 59% and 32% for Radiata E, Radiata R and Caribbean respectively. The consequence of 

32

choosing biased or random has a smaller impact on Caribbean than it does on the two Radiata 

resources. The greater impact is on the Radiata R resource. 

Table 3. Descriptive statistics of density and mechanical properties for the three resources, biased and 

random separated. 

Test position Resource Mean Std. Deviation N 

Biased Radiata E Dry density (kg/m 3 ) 484 51 517 









Random Radiata E Dry density (kg/m 3 ) 489 51 207 









The histograms in Figure 11and Figure 12 illustrate the test consequence of the position 

choice for the three resources on MOE and MOR. 

33

Figure 11. Biased position MOE (left), and random position MOE (right), for all resources. 

Figure 12. Histograms for biased sample MOR (left), and random position MOR (right), all resources. 

The bivariate Pearson's correlation coefficients between density, MOE and MOR are 

displayed in Table 4 for biased and random tests separately. The correlations between MOE 

34

and MOR with density are always better on random due to a wider span of values across the 

density range. Correlations between MOE and MOR are similar on biased and random for 

Radiata E and R whereas Caribbean has better correlations on random. 

Table 4. Bivariate Pearson's correlation coefficients between density, MOE and MOR, biased and 

random, all resources. 

Test 

Dry density 

Resource 

position 

(kg/m 3 MOE 

) (MPa) 

Biased Radiata E MOE (MPa) 0.68 






Random Radiata E MOE (MPa) 0.71 






All the correlations are significant at the 0.01 level (2-tailed). 

The latter observation on Caribbean can be explained by a larger dispersion of MOR values 

for MOE values above 12000 MPa (see Figure 13 and Figure 14). 

35

Figure 13. Biased test: MOE vs MOR, all resources. 

Figure 14. Random test: MOE vs MOR, all resources. 

36

Biased vs Random 

Figure 15, Figure 16 and Figure 17 display the correlations for each resource between MOR 

random and MOR biased tests performed on the same board at different locations. The red 

line represents the angle bisector of the graph. If selection of the weakest point on the board 

was perfect all the measurement points should be above this angle bisector line. For Radiata E 

resource the choice was generally good despite a few mistakes. In the case of Radiata R 

resource because some boards contained numerous large knots or defects it was sometimes 

difficult to nominate the weakest point unless without spending several minutes assessing 

each board. The result obtained for Caribbean is less easily explained because selection of the 

weakest point was apparently less difficult than Radiata R due to the proportion and sizes of 

the knots being somewhat smaller. Nevertheless it seems for some boards the nature or the 

type of the weakest knot or defect wasn’t assessed correctly before testing. 

37

Figure 15. Stiffness limited radiata pine, random vs biased MOR. 

Figure 16. Strength limited radiata pine, random vs biased MOR. 

38

Figure 17. Caribbean pine, random vs biased MOR. 

Position of the boards within the logs 

For all three resources the position of the board determined by distance from the pith within 

the log had only a weak correlation with the MOE or the MOR of the board when considering 

biased and random all together (Table 5). Caribbean displays the best correlation followed by 

Radiata E then Radiata R. 

Table 5. Distance from the pith vs static bending, all resources. 

Resource / distance from the pith - R 2 

MOE MOR 

Radiata E 0.13 0.08 

Radiata R 0.09 0.04 

Caribbean 0.21 0.11 

39

Individual NDT predictors for static MOE and MOR biased evaluation 

In the following section all the analyses presented are based on biased boards only according 

to the justification given in the earlier section Biased versus random position testing. 

Correlation analyses between the individual NDT parameters and the destructively measured 

properties were carried out. In this way, the coefficient of determination (R 2 ) of each 

measured property in linear regression as a predictor for the destructively determined 

properties was obtained. 

Our logic in presenting the results has been to start from the data collected chronologically as 

close as possible to the reference static bending i.e. the dry boards then present the results in 

reverse chronological order viz the green boards and finally for the logs. 

At dry mill level 

The R 2 values for each parameter are derived from the dried, dressed boards. The indicating 

properties obtained from the strength grading machines or equipment used are included as 

single parameters, even if some of them required more than one measurement. 

Radiata E resource 

Figure 18 and Figure 19 present the coefficient of determination obtained on Radiata E 

resource between the different individual predictors and the biased static bending MOE and 

MOR results. 

The best predictors for static MOE were achieved through the resonance method. We remind 

the reader that the dynamic MOE is based on the measurement combination of the vibration 

signal and the density. The flexion mode provides the higher coefficient of determination 

(Figure 18). Bernoulli’s model on the first resonance frequency and Timoshenko’s model both 

gave similar results. The second or third resonance frequencies with Bernoulli’s model gave a 

slightly lower level of correlation. 

The vibration MOE in compression mode is very close or even equivalent to the vibration 

MOE in flexion. The average MOE Metriguard profile on the reference destructive static test 

span (local measurement) provides a correlation 0.05 lower than the flexion resonance 

method. If the average Metriguard MOE profile is calculated on the full length of the board, 

the coefficient of determination is 0.04 less than the vibration MOE in compression. The 

minima MOE Metriguard provides a similar level of correlation to the Metriguard MOE 

board profile. One would expect that the local measurement over the test span should be 

better than the global measurement (on the full length of the board), however this is not 

necessarily the case. It is possible that the flatwise, 3-point bending as utilised in the 

Metriguard process combined with the knotty attributes of the stiffness-limited resource, 

produces this discrepancy. 

Interestingly the specific vibration MOE in compression as a predictor of static bending MOE 

provides an R 2 of 0.72 which is be acceptable to pre-sort the low stiffness boards with a very 

simple device able to measure only the board vibration. It may not be necessary to combine 

density as a parameter to achieve a satisfactory pre-grading result. 

The best WoodEye ® predictor for MOE is the KAR calculated with a triangular window 

(R 2 =0.31). 

40

Predictors MOE (MPa) MOR (MPa) 

MOE (MPa) 1.00 0.64 

Flex‐Dry‐MOET 

0.85 0.53 

Flex‐Dry‐MOE_1 

0.85 0.53 

Comp‐Dry‐MOE_1 

0.83 0.50 


0.82 0.50 


0.82 0.50 

Metri_avg Static Test 0.80 0.46 


0.80 0.48 


0.80 0.48 

Metri_avg board 0.79 0.47 


0.79 0.47 

Metri_min Static Test 0.78 0.50 

Metri_min board 0.77 0.52 

Flex‐Dry‐MOE_1 /Density 

0.75 0.41 

Comp‐Dry‐MOE_1/Density 

0.73 0.38 


0.65 0.38 


Flex‐Dry‐SBandwidth 

0.64 0.35 

Flex‐Dry‐SCGravity 

0.53 0.28 

Comp‐Dry‐SBandwidth 

0.47 0.24 

Dry Density (kg/m3) 0.46 0.37 

Comp‐Dry‐SSlope 

0.32 0.19 

WE_KAR_trgl_99 0.31 0.29 

WE_KAR_ext_99 0.29 0.27 

WE_KAR_ext_98 0.28 0.27 

Comp‐Dry‐SCGravity 

0.27 0.13 

WE_CTR_trgl_AC_M 0.26 0.22 

WE_KAR_trgl_M 0.26 0.24 

WE_CTR_AC_M 0.26 0.22 

WE_KAR_ext_M 0.26 0.23 

WE_CTR_trgl_AC_3 0.24 0.24 


Figure 18. R 2 of NDT measurements to destructively determined properties in bending for Radiata E 

resource on dry, dressed boards. Note: MOE R 2 sorted largest to smallest, lower limit = 0.2. 

The best prediction of MOR comes from static bending MOE which is appreciably above the 

second best predictor which is vibration MOE in flexion (Figure 18). Indeed the resonance 

method, whatever the configuration, and Metriguard give similar levels of prediction, around 

0.5. The MOEs calculated from higher resonance frequencies exhibit coefficient of 

determinations between 0.4 and 0.5. Because they are expressing the same physical property 

the levels of correlation with MOR are linked to those with static bending MOE. As a result, 

for simplification reasons, the Figure 18 displayed only the first resonance frequency MOEs 

in flexion and compression. Two signal descriptors, the spectral centre of gravity (R 2 =0.28) 

and the spectral bandwidth (R 2 =0.35), provide levels of prediction in the same range of the 

best optical scanner parameters. 

Dry density has an R 2 equal to 0.37, which is above the level of prediction of even the best 

WoodEye ® knot parameters which presents a coefficient of determination of 0.32. The fibre 

knot detection gives the best knot type prediction and the best mode of extraction is the 

maximum KAR. 

41




Flex‐Dry‐MOE_1 0.85 0.53 



Comp‐Dry‐MOE_1 0.83 0.50 



Dry Density (kg/m3) 0.46 0.37 

Fk_KAR_trgl_99 

Fk_KAR_ext_99 

Fk_KAR_ext_98 

WE_KAR_trgl_99 

WE_KAR_ext_99 

WE_KAR_ext_98 

Fk_CTR_trgl_AC_3 


Fk_KAR_trgl_M 

Fk_KAR_ext_M 

Fk_CTR_trgl_AC_M 

Fk_CTR_AC_M 

WE_CTR_trgl_AC_3 

WE_KAR_trgl_M 


WE_KAR_ext_M 

WE_CTR_trgl_AC_M 

WE_CTR_AC_M 

0.33 0.32 

0.30 0.30 

0.30 0.30 

0.31 0.29 

0.29 0.27 

0.28 0.27 

0.26 0.27 

0.26 0.26 

0.27 0.25 

0.26 0.25 

0.29 0.25 

0.29 0.25 

0.24 0.24 

0.26 0.24 

0.23 0.23 

0.26 0.23 

0.26 0.22 

0.26 0.22 


resource on dry, dressed boards. Only the first frequency vibration MOE for both compression and 

flexion are exhibited. Note: MOR R 2 sorted largest to smallest, lower limit = 0.2. 

Figure 20 displays the high correlation between Metriguard average MOE on the board and 

vibration MOE in compression. The coefficients of determination are very high because the 

physical principles of measurement of these two methods are both established on strength of 

material principles and provide a direct measure of MOE. Nevertheless, they are based on 

different loading modes which explains the slight discrepancy observed through the linear 

regression equation (constant different from 0) and the residual error (dispersion around the 

regression line). Three factors can explain the observed difference of the experimental data: 

• Metriguard measures a long-span flat-wise average MOE whereas the vibration 

compression MOE is measured through a uniform stress wave across the whole 

section. 

• The tested span is shorter for Metriguard because the measurement span starts and 

finishes at 80 cm from both ends whereas the resonance method applies a stress on full 

length of the board. 

• The longitudinal resonance method induces a compression-traction stress whereas 

Metriguard system induces a three point bending stress. The latter provokes a shear 

effect which should lead to a lower measured MOE. One should note that the constant 

of the regression equation is the same order of magnitude as the shear modulus of 

elasticity in the longitudinal-radial plane of the wood. Indeed Metriguard provides a 

higher MOE than vibration MOE in compression. There is no obvious explanation to 

this phenomenon. 

42

Metriguard Average board (MPa) 

20000 

18000 

16000 

14000 

12000 

10000 

8000 

6000 

4000 

2000 

0 

y = x + 1590 

R² = 0.96 

0 2000 4000 6000 8000 10000 12000 14000 16000 18000 20000 

Vibration Compression MOE_1 (MPa) 

Figure 20. Linear regression between MOE measured by resonance method in compression and 

average MOE given by Metriguard for Radiata E resource on dry, dressed boards. 

Radiata R resource 

Figure 21 and Figure 23 present the coefficient of determination obtained on Radiata R 

resource between the different individual predictors and the biased static bending MOE and 

MOR results. 

For this resource the X-ray scanner LHG was also included in the set of methods evaluated. 

The Metriguard HCLT provided the strongest correlation with static MOE, using the 

minimum MOE from the profile on the static bending test span (its local measurement). 

The LHG predicted strength, which combines the previous parameter and X-ray knot 

descriptors, provided the second highest correlation as expected, due to the combination of 

Metriguard (minimum MOE from the profile on the static bending test span) and X-ray 

parameters. Vibration MOE in flexion, compression and flexion using Timoshenko’s model, 

all deliver comparable quality of prediction. These predictors are closely followed by the 

average static test (local measurement). Surprisingly, this local parameter (matching the test 

span for the reference test) did not provide a stronger correlation than the global 

measurements. 

Taken as a whole, the best predictors for the R limited resource have an average R 2 difference 

of 0.15 with Radiata E resource. This reflects the knotty nature of this resource which 

significantly impacts the results provided by individual or simple combined parameters. 

Specific knot information should be qualitatively and quantitatively added to a prediction 

model in order to improve the predictions. 

As for Radiata E, the specific MOE correlation for the Radiata R was in the same magnitude 

of order to the best correlations, indicating that simple vibration systems may form the basis 

of an effective pre-grading tool. 

43

For this resource, the R 2 for MOEs calculated using higher resonance frequencies were 0.05 

lower than the first frequency. The Metriguard average MOE displays R 2 significantly lower 

than other Metriguard and vibration parameters. Despite the characteristic knottiness of the 

Radiata R resource, the knot parameters as measured by WoodEye ® didn’t provide as good 

indications for stiffness as for the Radiata E resource. 




LHG predicted Strength 0.67 0.53 


0.66 0.28 


0.65 0.28 


0.65 0.29 



0.63 0.26 








0.61 0.26 

0.61 0.26 

0.60 0.24 

0.59 0.23 

0.59 0.26 

0.57 0.25 




0.49 0.16 


0.32 0.08 


0.32 0.16 


0.29 0.15 

Dry Density 0.28 0.12 

WE_KAR_trgl_M 0.27 0.30 

WE_KAR_ext_M 0.27 0.30 

WE_KAR_trgl_99 0.26 0.36 

WE_KAR_ext_99 0.24 0.35 

WE_KAR_ext_98 0.24 0.35 


0.23 0.10 


0.23 0.11 


0.23 0.12 

LHG KAR 0.21 0.25 


Figure 21. R 2 of NDT-measurements to destructively determined properties in bending for Radiata R 

resource on dry, dressed boards. Note: MOE R 2 sorted largest to smallest, lower limit = 0.2. 

For MOR prediction, the MOEs calculated from higher resonance frequencies exhibit 

coefficient of determinations between 0.1 and 0.3. Because they are expressing the same 

physical property the levels of correlation with MOR are linked to those with static bending 

MOE. As a result, for simplification reasons, Figure 23 displays only the first resonance 

frequency MOEs in flexion and compression. Three signal descriptors, the spectral centre of 

gravity (R 2 =0.1), the spectral slope (R 2 =0.15) and the spectral bandwidth (R 2 =0.16), provide a 

level of prediction significantly lower than the best optical scanner parameters (R 2 between 

0.3 and 0.4). 

LHG predicted strength provided R 2 of 0.53, the best result from all MOR predictors (Figure 

23). As depicted in Figure 22 below, only the local measurements provided reasonable 

correlations for predictions, with Metriguard HCLT (minimum MOE) providing the best 

correlation to the static bending MOE. The test span information provides a significantly 

superior prediction of strength (R 2 =0.47). 

44

The R 2 for Metriguard minimum of the board was 0.1 lower than the Metriguard Static test 

span result (0.37 vs 0.47 respectively). This implies that the minimum for the board is not the 

best predictor from the Metriguard MOE profile for this resource. In the case of the Radiata E 

resource, these two parameters provided a similar level of correlation. 

From the optical scanner information, WoodEye ® KAR provided the best correlation with 

MOR, surpassing LHG X-ray by 0.1 (0.36 vs 0.25 respectively). The Metriguard minimum 

static test contains the majority of the information and combining with LHG only improves by 

a magnitude of 0.06. 

Of the various knot types detected by WoodEye ® , fibre knots provide the best correlation, 

with a similar level of correlation to the combined knot silhouette. 

The Metriguard average and the vibration compression MOE both provide the same level of 

correlation. 

45




LHG predicted Strength 0.67 0.53 




0.26 0.36 

WE_KAR_ext_99 

0.24 0.35 

WE_KAR_ext_98 

0.24 0.35 


0.25 0.34 


0.23 0.34 

Fk_KAR_ext_99 

0.21 0.32 

Fk_KAR_ext_98 

0.21 0.32 

WE_KAR_ext_M 

0.27 0.30 

WE_KAR_trgl_M 

0.27 0.30 


0.21 0.30 


0.21 0.29 


0.20 0.29 


0.19 0.29 

Fk_KAR_trgl_M 

0.25 0.29 

Fk_KAR_ext_M 

0.24 0.28 




WE_KAR_trgl_M 

0.26 0.26 

LHG KAR 0.21 0.25 

LHG weighted KAR 

0.19 0.23 


WE_KAR_int_99 

0.16 0.21 

WE_KAR_int_M 

0.18 0.19 

Fk_CTR_trgl_AC_M 

0.18 0.19 

Fk_CTR_AC_M 

0.18 0.19 


0.18 0.19 

Fk_KAR_int_99 

0.13 0.18 

WE_CTR_AC_M 

0.18 0.18 

Fk_KAR_int_M 

0.16 0.18 

LHG I ratio 0.12 0.16 

Figure 23. R 2 of individual NDT-measurements to destructively determined properties in bending for 

Radiata R resource on dried, dressed boards. Only the first frequency vibration MOE for both 

compression and flexion are exhibited. Note: MOR R 2 sorted largest to smallest, lower limit = LHG I 

ratio. 

Again we found a very good correlation for Metriguard average MOE with vibration 

compression MOE (R 2 = 0.90). However, due to the knot characteristics of the strength 

limited resource and subsequent outliers, as seen in Figure 24 below, the correlation is lower 

than was achieved for the stiffness limited resource. 

46

Metrigiard Average Board (MPa) 

20000 

18000 

16000 

14000 

12000 

10000 

8000 

6000 

4000 

2000 

0 

y = 0.9471x + 1912 

R² = 0.90 

0 2000 4000 6000 8000 10000 12000 14000 16000 18000 20000 



average MOE given by Metriguard for Radiata R resource on dried, dressed boards. 

Caribbean resource 

The levels of correlation observed for the Caribbean biased static bending MOE are slightly 

lower than achieved for Radiata E. For this resource the Metriguard provided the best 

prediction of MOE (R 2 =0.84). As expected, the Metriguard average biased static test was 

better than the Metriguard average board result. In contrast to the Radiata E resource, the 

vibration MOE correlations were lower than for Metriguard. 

Vibration MOE (compression) without combining density still achieved a reasonable 

correlation (R 2 = 0.62). As for the other parameters, the vibration descriptors provided a range 

of R 2 from 0.30 to 0.40 for the best correlations. 

47






0.79 0.34 


0.79 0.33 


0.78 0.31 



0.77 0.32 


0.77 0.32 


0.76 0.31 


0.76 0.31 


0.76 0.32 


0.73 0.30 



0.62 0.23 


0.53 0.23 


0.50 0.20 



0.49 0.18 


0.46 0.18 


0.43 0.16 


0.39 0.15 


0.32 0.12 

Comp‐Dry‐MSNrjRatio_4 

0.24 0.09 

Comp‐Dry‐MSRatioF_3 

0.21 0.07 

Comp‐Dry‐MSPowerF_4 

0.20 0.10 


Caribbean resource on dry, dressed boards. Note: MOE R 2 sorted largest to smallest, lower limit = 0.2. 

For MOR prediction, the MOEs calculated from higher resonance frequencies exhibit 

coefficient of determinations between 0.2 and 0.3. Because they are expressing the same 

physical property the levels of correlation with MOR are linked to those with static bending 

MOE. As a result, for simplification reasons, Figure 25 displays only the first resonance 

frequency MOEs in flexion and compression. Three signal descriptors, the spectral centre of 

gravity, the spectral slope and the spectral bandwidth, provided coefficient of determinations 

between 0.1 and 0.2. They are slightly lower than the best optical scanner parameters (R 2 

between 0.15 and 0.25). 

As for the MOE prediction, the best predictor of MOR for the Caribbean resource was 

provided by the Metriguard, local parameters (R 2 =0.37, Figure 26). This is significantly lower 

than for the Radiata E and Radiata R (both R 2 =0.53). Additionally, WoodEye ® provides a 

lower level of prediction in the case of Caribbean pine (R 2 = 0.24, best result for Caribbean 

compared to R 2 = 0.36 for Radiata R and R 2 =0.32 for Radiata E). The Caribbean pine knot 

attributes are generally captured by WoodEye ® scanners as ‘black knots’, whereas this feature 

is less frequently detected for the radiata resources, whose knot attributes are mostly detected 

as fibre knots. This difference in knot attributes/descriptors may explain the low level of 

correlation for MOR. This may be due to the incidence of loose knots in the Caribbean 

material. 

48










WE_KAR_ext_99 

WE_KAR_ext_98 




WE_KAR_ext_M 

WE_CTR_AC_5 

WE_KAR_trgl_M 

0.04 0.24 

0.04 0.24 

0.05 0.22 

0.04 0.20 

0.05 0.19 

0.03 0.18 

0.03 0.16 

0.03 0.16 

Bk_CTR_trgl_AC_3 0.06 0.16 

Fk_KAR_ext_99 

0.01 0.15 

Fk_KAR_ext_98 

0.01 0.15 

Bk_CTR_trgl_AC_5 0.05 0.15 

Bk_CTR_AC_5 0.05 0.14 


0.01 0.14 


0.02 0.13 


0.02 0.13 

Bk_CTR_AC_M 0.05 0.12 

Bk_CTR_trgl_AC_M 0.05 0.12 

Fk_CTR_AC_5 

0.01 0.12 

Bk_KAR_trgl_99 0.02 0.11 

Fk_KAR_ext_M 

0.01 0.11 


0.03 0.11 


Caribbean resource on dry, dressed boards. Only the first frequency vibration MOE for both 

compression and flexion are exhibited. Note: MOR R 2 sorted largest to smallest, lower limit = 0.1. 

As was found for the two radiata resources, a high correlation was found between the 

Metriguard average board MOE and vibration compression MOE (R 2 =0.96, Figure 27). 

49

Metrigiard Average Board (MPa) 

25000 

20000 

15000 

10000 

5000 

0 

y = 0.95x + 1294 

R² = 0.96 

0 5000 10000 15000 20000 25000 



average MOE given by Metriguard for Caribbean resource on dry, dressed boards. 

At green mill level 


The vibration MOE in compression provides the best correlation with static bending MOE 

and MOR (R 2 =0.77 and R 2 =0.47 respectively) as depicted in Figure 28. MOE from flexion 

vibration displays a slightly lower level of correlation (R 2 =0.75 and R 2 =0.45 respectively). 

The best predictor on dry dressed boards is from flexion vibration followed by compression 

vibration. They are calculated both from the first frequency with R 2 =0.85 and R 2 =0.83 for 

MOE and R 2 =0.53 and R 2 =0.50 for MOR respectively. These are better than the green board 

prediction by a magnitude of 0.05 to 0.08. The specific compression vibration MOE as a 

single predictor didn’t provide a level of correlation suitable for sorting purposes. 

50



Comp‐Green‐MOE_1 

0.77 0.47 


0.76 0.45 

Flex‐Green‐MOE_3 

0.75 0.45 

Flex‐Green‐MOET 

0.74 0.45 


0.74 0.45 


0.74 0.44 

Thumper_MOE (MPa) 0.74 0.44 


0.73 0.45 


0.72 0.44 


0.70 0.41 


Flex‐Green‐FacQ_4 

0.36 0.23 

Flex‐Green‐SSlope 

0.29 0.17 

Green Density 0.27 0.19 

Flex‐Green‐MOE_1 /Density 

0.26 0.13 

Comp‐Green‐MOE_1/Density 

0.25 0.12 

Gamma_Density(Kg/m3) 0.23 0.17 

Comp‐Green‐SSlope 

0.22 0.13 

Flex‐Green‐SBandwidth 

0.20 0.11 

Comp‐Green‐SBandwidth 

0.17 0.08 

Flex‐Green‐SCGravity 

0.13 0.07 

Flex‐Green‐R2T 

0.12 0.10 


resource on green sawn boards. Note: MOE R 2 sorted largest to smallest, lower limit = 0.1. 

This difference between green board and dry board prediction is statistically significant but it 

is still probably low enough to consider a pre-grading system for sorting the low stiffness 

green boards. This was partially done in one mill with the Thumper system which provides a 

high level of prediction equivalent to the one obtained with Bing® system. Indeed these two 

systems should provide the same MOE values. 

Figure 29 depicts the correlation between the green board compression MOE from the first 

frequency between Bing ® and Thumper system. The correlation is not as high as expected and 

the residual error is quite significant. This discrepancy may be explained by the effect of 

uncontrolled drying on green density which occurred between the two set of measurements. 

Bing ® measurements were performed three weeks after Thumper trial. The densities observed 

on the latter was on average 6% higher than the densities when the boards were measured for 

Bing ® measurements with an important dispersion due to the position of the board in the stack 

(R 2 =0.86 between green density measured by direct weighing and gamma ray system). 

Moreover the acoustic equipment can differ, i.e. settings (support effect, dimensions 

measurements etc.) and extraction algorithms (peak detection) which can lead to some small 

differences in the MOE calculation. 

51

Thumper MOE (MPa) 

18000 

16000 

14000 

12000 

10000 

8000 

6000 

4000 

2000 

0 

y = 1.02x 

R² = 0.89 

0 2000 4000 6000 8000 10000 12000 14000 16000 18000 

Green compression Bing MOE_1 (MPa) 

Figure 29. Linear regression between MOE from resonance method in compression measured by off 

line Bing ® and by in-line Thumper systems, for Radiata E resource on green sawn boards. 


Similarly to Radiata E resource, the vibration MOE in compression provides the best 

correlation with static bending MOE and MOR (R 2 =0.58 and R 2 =0.25 respectively) as 

depicted in Figure 30. MOE from flexion vibration displays a marginally lower level of 

correlation (R 2 =0.57 and R 2 =0.25 respectively). 

For this resource the level of correlation of the best predictors is around 0.2 lower than 

Radiata E. This difference is higher than the difference observed on dry boards (around 0.15). 

The drying process certainly plays a role in this difference since the differential shrinkage 

from the knotty area may induce cracks and lack of adherence which may contribute to 

blurring the correlations between green and dry products. 

Similar to Radiata E resource, the specific compression vibration MOE displays a very low 

prediction of static bending MOE and MOR. 

52

Predictors MOE (Mpa) MOR (Mpa) 

MOE (Mpa) 1.00 0.56 


0.58 0.25 


0.57 0.25 


0.57 0.24 

MOR (Mpa) 0.56 1.00 


0.56 0.25 


0.54 0.22 


0.53 0.23 


0.53 0.23 


0.52 0.21 


0.51 0.21 


0.49 0.19 

Flex‐Green‐MOE_1/Density 

0.27 0.08 


0.27 0.08 


0.27 0.15 


0.19 0.09 


0.16 0.09 


0.16 0.06 


0.14 0.09 

Flex‐Green‐MSPowerF_2 

0.12 0.03 

Flex‐Green‐MSPower_1 

0.12 0.02 


0.12 0.08 

Flex‐Green‐Q_2 

0.10 0.05 

Flex‐Green‐MSPowerF_3 

0.10 0.03 

Flex‐Green‐MSPower_2 

0.09 0.01 

Comp‐Green‐Q_4 

0.09 0.04 


Figure 30. R 2 of NDT measurements to destructively determined properties in bending for Radiata R 

resource on green sawn boards. Note: MOE R 2 sorted largest to smallest, lower limit given by the 

green density. 


For the Caribbean resource, the best correlations are provided by flexion vibration which is 

marginally better than compression vibration. Contrasting with the other resources the best 

predictors are slightly improved from the use of higher resonance frequencies. The 

compression wood occurrence coupled together with high resin content may explain this 

observation already described in the static bending results paragraph above. 

The vibration MOE in flexion provides the best correlation with static bending MOE and 

MOR (R 2 =0.75 and R 2 =0.31 respectively) as depicted in Figure 31. MOE from compression 

vibration displays a marginally lower level of correlation (R 2 =0.74 and R 2 =0.30 respectively). 

Apart from the vibration MOE extracted parameters, the signal descriptors provide a higher 

static bending MOE prediction than the other resources but not for MOR prediction. 

53




0.75 0.31 


0.75 0.32 


0.74 0.32 


0.74 0.31 


0.74 0.30 


0.73 0.30 


0.73 0.30 


0.73 0.30 


0.73 0.30 


0.72 0.30 


0.63 0.25 

Flex‐Green‐MOE_1/Density 

0.58 0.23 



0.48 0.18 


0.43 0.17 


0.43 0.16 


0.41 0.16 


0.37 0.13 

Comp‐Green‐SCGravity 

0.31 0.13 

Flex‐Green‐Q_2 

0.29 0.16 

Comp‐Green‐Alpha_1 

0.27 0.09 

Comp‐Green‐MSNrjRatio_3 

0.26 0.09 

Comp‐Green‐MSPower_3 

0.25 0.10 

Comp‐Green‐MSRatioF_3 

0.25 0.10 

Comp‐Green‐Q_2 

0.25 0.10 

Flex‐Green‐Alpha_2 

0.23 0.09 

Comp‐Green‐MSRatioF_4 

0.23 0.11 


0.22 0.09 

Comp‐Green‐MSNrjRatio_4 

0.22 0.10 


Figure 31. R 2 of NDT measurements to destructively determined properties in bending for Caribbean 

resource on green sawn boards. Note: MOE R 2 sorted largest to smallest, lower limit given by the 

green density. 

54

Combinations of NDT predictors for static MOE and MOR evaluation 

We remind the reader that only the results from the biased static bending tests are analyzed. 

The compression vibration system is more practical and easier to carry out in the mill than a 

flexion vibration because of the boundary conditions robustness and the fewer complementary 

data used. Moreover from the previous chapter we have noticed that the compression 

vibration and the flexion vibration provide comparable levels of prediction. For these reasons 

the following analysis will only focus on compression vibration data in combination with the 

other systems tested. 

In this chapter, two statistical approaches are used: the Multiple Linear Regression (MLR) 

and the Partial Least Squares (PLS). 

The variable selection for MLR was performed through a stepwise process: the selection 

process starts by adding the variable with the largest contribution to the model (the criterion 

used is Student's t statistic). If a second variable is such that the probability associated with its 

t is less than the "Probability for entry", it is added to the model. The same protocol applies 

for a third variable. After the third variable is added, the impact of removing each variable 

present in the model after it has been added is evaluated (still using the t statistic). If the 

probability is greater than the "Probability of removal", the variable is removed. The 

procedure continues until no more variables can be added or removed. The collinearity (or 

multicollinearity) problem was prevented by limiting the number of variables introduced in 

the model through a condition index threshold. Heteroscedasticity (non-homogeneous 

variance) was checked. If the model was subject to an heteroscedasticity problem an 

appropriate data transformation was applied. This was often the case for MOR prediction 

where a logarithm-transform was performed. 

PLS attempts to find factors which both capture variance and achieve correlation. PLS 

attempts to maximize covariance and this is the explicit objective of the simple PLS 

(SIMPLS) algorithm used in this study. Two rules of thumb with regard to selecting the 

optimal number of factors were applied, namely: 

• to only choose additional factors when the RMSEC improves by at least 2%, and 

• to choose as few factors (called Latent Variables) as possible. 

When developing the models an auto-scaling process was used (mean-centring followed by 

dividing variable by the standard) and no cross-validation performed. A log transformation 

was systematically applied on MOR values. 


The coefficient of determination for predicting the static bending MOE from the best 

combinations of methods or parameters on dry boards ranges from 0.79 to 0.86 (Table 6). 

Using average MOE and minimum MOE together from Metriguard profiles in MLR doesn’t 

improve the correlation. Only the profile average provides the best prediction from this 

equipment on Radiata E resource. 

From the 35 variables extracted from each vibration spectrum only the first frequency 

vibration MOE was selected as a suitable explanatory variable in MLR. This fact is confirmed 

by the PLS model which displays the same level of prediction despite its complexity. The 

55

only advantage to use a PLS approach could be the robustness of the model since it uses 

several variables to perform the prediction. 

Adding WoodEye ® and/or Metriguard information to vibration measurement only improved 

the level of prediction by a maximum of 0.03. This observation indicates that the local 

information provided by WoodEye ® or Metriguard marginally improves the static bending 

MOE prediction. 

For the green boards, the static bending MOE prediction level is below that observed for dry 

boards by an average of only 0.06. If an optical or equivalent system can be used on green 

boards with the same quality of information as provided on dry boards, then there is a small 

improvement when combined with vibration using PLS on all parameters (+0.02). 

Table 6. R 2 of combined NDT measurements to destructively determined properties in bending for 

Radiata E resource on green sawn boards. Note that the number listed beside PLS indicates the 

number of factors used in the models. Each colour corresponds to a different set of equipment or wood 

moisture content. 

Methods Reg. Parameters R2 x 100 


Vibration GREEN 

Vibration GREEN + 

WoodEye 

MLR/P 

LS 3 

56 

Parameters 

R 2 x 100 


MOE_1 77 MOE_1 48 

PLS 3 all 79 all 55 

Metriguard Board MLR Avg_Brd 79 Mini_Brd 52 

Metriguard Test Span MLR Avg_Span 80 Mini_Span 52 

Vibration MLR MOE_1 83 

MOE_1 + 

MOE_1/density + 

Alpha_2 

Vibration PLS 4 all 83 all 57 

Vibration + Metriguard MLR 


Mini_Span 

53 

85 Mini_Span + MOE_1 54 

Vibration + Metriguard PLS 4 all 85 all 61 

Vibration + WoodEye MLR 


Fk_KAR-m 

84 MOE_1 + KAR-Ext 55 

Vibration + WoodEye PLS 3 all 84 all 62 

Vibration + WoodEye + 

Metriguard 


Metriguard 

MLR 


Mini-Span 

85 

Mini_Brd + Kar-Ext + 

MOE_1 

PLS 3 all 86 all (PLS 4) 64 

When considering the MOR predictions, MLR provided an R 2 of 0.52 up to 0.57, with the 

best combination being the combination of three systems: vibration, WoodEye ® and 

Metriguard. This level of prediction is achieved by using only the vibration methods in 

combination with a PLS model. In combination with Metriguard or WoodEye ® , the prediction 

increases to 0.62. The best combination is achieved with the three systems combined where 

an R 2 of 0.64 resulted through PLS approach. 

57

The first frequency vibration MOE alone provided an R 2 of 0.48 which wasn’t improved by 

using MLR or PLS. Interestingly, using PLS on vibration parameters for green boards 

combined with an optical scanner achieved a prediction level as good as MLR on dry boards 

using a combination of vibration and Metriguard or WoodEye ® (R 2 =0.55). 


On this resource, Metriguard only achieved an R 2 of 0.70 for static bending MOE prediction, 

which is 0.1 below the other two resources. The compression vibration system, whatever the 

statistical method, doesn’t provide a better correlation; however, combining vibration 

measurements with local information obtained from WoodEye ® or Metriguard, significantly 

improves the prediction (up to 0.77). As shown in Table 7 there is only marginal 

improvement by combining all three methods. 

Combining LHG with vibration doesn’t significantly improve the MOE prediction when 

compared to Metriguard alone. Metriguard combined with LHG provided an R 2 between 

Metriguard and vibration plus Metriguard or WoodEye ® . The use of PLS method of 

regression doesn’t improve MOE prediction compared to using MLR using 2 or 3 parameters. 

On green boards when using PLS method, vibration alone indicates a level of correlation 

equivalent to the Metriguard average MOE profile on the board. If an optical or equivalent 

system can be used on green boards with the same quality of information as provided on dry 

boards, then there is a large improvement when combined with vibration using PLS on all 

parameters (approximately +0.1). 

Contrary to Radiata E, these results prove that the characteristic knotty nature of his resource 

heavily impacts the MOE, since when local parameters are combined with global 

measurements the prediction improvement is substantial. 

57


Radiata R resource on green sawn boards. Note that the number listed beside PLS indicates the 

number of factors used in the models. Each colour corresponds to a different set of equipment or wood 

moisture content. 

Methods Reg. Parameters 

R 2 x 100 


Parameters 

R 2 x 100 


Vibration GREEN MLR MOE_1 59 MOE_1 26 

Vibration GREEN PLS 4 all 62 all 33 


WoodEye 

PLS3 all 71 all 55 


Metriguard Test span MLR Avg_Span 70 Mini_Span 47 

Vibration MLR 


Mspower 

58 

67 


MOE_1/density + 

Alpha_2 


Vibration + Metriguard MLR 


Mini_Span 

77 

Mini_Span + 

MSpower2 

48 




KAR_Ext 

75 MOE_1 + KAR_Ext 49 

Vibration + WoodEye PLS 4 all 76 all 59 


Metriguard 

MLR 


KAR_Ext + 

Mini_Span 

78 

MOE_1 + KAR_Ext 

+ MSPower2 

57 


Metriguard 


Vibration + LHG 

(without Strength 

prediction) 

MLR 


Weighted_KAR 

70 


Weighted_KAR + 

MSpower2+ LHG Iratio 

43 

LHG strength prediction 

Vibration + LHG 

LR 

Indicative 

property 

67 Indicative property 53 

(without Strength 

prediction) 


Metriguard + LHG MLR 

Mini_Span + 

Avg_brd 

73 

Mini_Span + LHG I 

Ratio 

51 

Metriguard + LHG PLS 

not enough 

variables 

- not enough variables - 

The level of prediction for strength is rather low for most methods. The trend is the same as 

for MOE, in that when local information and global information are combined the prediction 

is significantly improved. Interestingly the PLS regression using data from vibration and 

WoodEye ® systems provides one of the highest prediction of strength (R 2 =0.59). The R 2 for 

LHG strength prediction is 0.53 which is the parameter currently used by the processor with 

this equipment. With additional optical equipment this may be improved significantly. 


Similarly to Radiata E resource, the coefficient of determination for predicting the static 

bending MOE from the best combinations of methods or parameters on dry boards, ranges 

from 0.79 to 0.87 (Table 8). Using average MOE from Metriguard profiles from the static 

31

ending span test significantly improves the correlation compared to using the average of the 

board. 

Vibration system alone provides a level of prediction equivalent to Metriguard from the 

average MOE on the board. These two systems combine together doesn’t improve the static 

bending MOE. The average MOE from Metriguard profiles on static bending span remains 

the only reliable predictor since the MLR stepwise process selects only this parameter to 

perform the prediction. 

Adding WoodEye ® and/or Metriguard information to vibration measurement only improves 

the level of prediction by a maximum of 0.05. This observation indicates that the local 

information provided by WoodEye ® or Metriguard improves the static bending MOE 

prediction. The best prediction (R 2 =0.87) was achieved by using the WoodEye ® (KAR outer 

third zone) and Metriguard (average MOE on test span) combined, without any measurable 

improvement by adding vibration to the combination. The best predictions are close to the 

maximum feasible level of correlation for MOE. Indeed the coefficient of determination 

reaches an asymptote thus it is very hard to improve the correlation as the measurement error 

is the main limiting factor. 

For the green boards, the static bending MOE prediction level is below that observed for dry 

boards by an average of 0.08. If an optical or equivalent system can be used on green boards 

with the same quality of information as provided on dry boards, then there is a small 

improvement when combined with vibration using PLS on all parameters (+0.02). 


Caribbean resource on green sawn boards. Note: MOE R 2 sorted largest to smallest, lower limit given 

by the green density. Note that the number listed beside PLS indicates the number of factors used in 

the models. Each colour corresponds to a different set of equipment or wood moisture content. 

Methods Reg. Parameters 

R 2 x 100 


Parameters 

R 2 x 100 


Vibration GREEN MLR MOE_1 73 MOE_1 31 

Vibration GREEN PLS 3 all 75 all 34 


WoodEye 

PLS 2 all 77 all (PLS 3) 55 


Metriguard Test Span MLR Avg_Span 84 Mini_Span 36 

Vibration MLR MOE_1 + FacQ1 78 MOE_1 32 


Vibration + Metriguard MLR Avg_Span 84 Mini_Span 36 




Fk_KAR-Ext 

79 


KAR-Ext 

46 

Vibration + WoodEye PLS 4 all 84 All (PLS 3) 55 


Metriguard 

MLR 

Avg_Span + KAR- 

Ext 

87 

Mini_Span+ 

Kar-Ext 

49 


Metriguard 


Despite the Caribbean resource having a similar range in the results for MOE prediction as for 

the Radiata E resource (R 2 =0.79 to 0.87 and 0.79 to 0.86 respectively), the level of accuracy 

was not as strong for strength prediction (R 2= 0.30 to 0.55 and 0.52 to 0.64). The poor level of 

59

strength prediction for Caribbean pine using current systems may be attributable to a feature 

of the knots in the resource tested, as discussed in the earlier section on individual NDT 

predictors. 

The same level of strength prediction achieved on dry boards was achieved on green boards 

(R 2 =0.55). This calculation was determined by the combined vibration (green boards) and 

WoodEye ® (dry boards) data, and assumes that the optical data can be captured at the green 

stage with equivalent accuracy to that captured on dry boards. The use of PLS regression 

provides the best predictions. 

Prediction of static bending MOE and MOR from logs 

This section uses the average properties of all the boards from the log as the basis for the 

correlations. It does not use log property as a predictor of individual board properties. A 

summary of results is provided in Table 9. 

Radiata E 

The MOE velocity was calculated using Equation 1, combination of log density data and 

velocity (TOF) measurements in the longitudinal axes. When correlated with the average 

static MOE levels of prediction of R 2 of 0.64 for MOE and 0.33 for MOR were provided. 

Slightly better correlations were observed when using velocity alone (TOF and distance 

between sensors) to predict MOE and MOR. These are lower than the prediction provided by 

vibration analyses (R 2 =0.70 and 0.45 respectively). 

The MLR combining compression and flexion tests increases the coefficient of determination 

from 0.70 to 0.76 for MOE and 0.45 to 0.61 for MOR, significant improvements, especially in 

the case of MOR prediction. However, this combination of measurements would require 

additional systems to enable measurements in the two planes. When developing the stepwise 

MLR to combine compression and flexion parameters, the condition index was close to the 

threshold, indicating a potential collinearity problem. The quality of this prediction needs to 

be considered with care until a larger population of logs can be tested for verification. 

Radiata R 

The velocity MOE measurements provided an R 2 of 0.50 for static MOE and 0.34 for MOR. 

The MOE prediction from logs for this resource is quite significantly lower compared to the 

Radiata E prediction (-0.14). The MOR predictions were similar for both resources. 

Noticeably better correlations were observed when using velocity alone (TOF and distance 

between sensors) to predict MOE and MOR. 

Vibration analysis provides better predictions compared to velocity measurements, with the 

best result using MLR with specific compression MOE (R 2 =0.75). When predicting strength 

using the same parameter from the log tests, the R 2 achieved was 0.43. These results are 

interesting due to the fact that only the first frequency measurement was required and 

therefore the density data wasn’t necessary to improve the result. This has positive 

implications for in-mill engineering and logistics in that weighing and log measuring systems 

may not be necessary for MOE prediction, but a larger log population should be tested to 

verify these results. 

60

Caribbean 

The velocity MOE measurements provided an R 2 of 0.56 for static MOE and 0.27 for MOR. 

The MOE prediction from logs for this resource is between the two radiata resources. The 

MOR prediction was significantly lower than the other material (-0.06). Conversely to the two 

other resources, markedly poorer correlations were observed when using velocity alone (TOF 

and distance between sensors) to predict MOE and MOR. 

As with radiata pine, vibration analyses provided better predictions compared to velocity 

measurements. 

Table 9. R 2 of combined NDT measurements to log average, destructively determined properties in 

bending for all resources. Note that the number listed beside PLS indicates the number of factors used 

in the models. Each colour corresponds to a different set of measurement. 

Resource 

Biased + 

Random 

Methods Reg. 

type 

Parameters 

61 

R 2 x 100 


Parameters 

R 2 x 

100 


Velocity 

(TOF) 


LR Velocity 65 Velocity 36 

Velocity 

(TOF) 

LR Velocity_MOE 64 Velocity_MOE 33 

Radiata E 

(n= 34) 

Specific 


LR 

Comp-MOE_1 

/density 

65 Comp-MOE_1 /density 43 

Comp LR Comp_MOE_1 70 comp-SBandwidth 45 

Comp + 

Flex 

MLR 

Comp_MOE_1, 

Ratio (Comp/Flex) 

76 

comp-MOE_1, Flex- 

MSPower_0 

61 

Velocity 

(TOF) 



Velocity LR Velocity_MOE 50 Velocity_MOE 34 

Radiata R 

(n= 27) 

(TOF) 

Specific 


LR 


/density 


Comp MLR 


/density 


Velocity 

(TOF) 



Velocity 

(TOF) 

LR Velocity_MOE 56 Velocity_MOE 27 

Caribbean 

(n= 106) 

Specific 


LR 


/density 


Comp MLR 

Comp-MOE_4, 

Comp-Alpha_2 

75 

Comp-MOE_3, Comp- 

Alpha_2 

37 

Comp PLS 3 all 78 all 44 

PLS analysis was able to be used on the Caribbean resource due to the higher number of logs 

in the study. As with green boards, the improvement for MOE prediction using the PLS 

method for Caribbean logs is slight, but for MOR it is a significant improvement. 

The reader should note that when using the data for all the boards per log, instead of the 

average of all boards, for the predicted static bending MOE and MOR, the coefficient of 

determination decreases by approximately half.

Practical grading into MGP grades 

Using best vibration prediction for logs 

For each resource, the best prediction equation from the resonance method was used to 

simulate the MGP grade recoveries. 

The removal of logs was based on a predicted threshold for MOE and MOR. The strategy was 

to target loss of no more than 5% of the potential MGP10 recovery through removal of logs. 

Plots for the most promising logs in each resource are compared in Figure 32 below which 

displays the percentage loss for each MGP grade and non-structural material. Because the 

MGP10 losses are standardised between the three resources, there is little difference in the 

MGP10 recovery. 

For Radiata E, 13.5% of the logs were removed using the 5% threshold for MGP10. 

Consequently, 32.1% of the non-structural material, 3.4% MGP10, and 0% for MGP12, were 

removed. 

For Radiata R, 4.6% of the logs were removed using the 5% threshold for MGP10. 

Consequently, 8.7% of the non-structural material, 3.3% MGP10, and 0% for MGP12 and 

MGP15, were removed. 

For Caribbean, 5.9% of the logs were removed using the 5% threshold for MGP10. 

Consequently, 13.3% of the non-structural material, 4.7% MGP10, 0% for MGP12, and 1.2% 

MGP15 were removed. 

The MGP10 recovery of all three resources is similar as it was a key criterion for setting the 

threshold for rejection of logs. 

For each plot, the non-structural (NS) had the lowest recovery in the remaining pieces. The 

implications derived from these results will vary between processors and resource 

characteristics. 

62

100% 

90% 

80% 

70% 

60% 

50% 

40% 

30% 

20% 

10% 

0% 

NS MGP10 MGP12 MGP15 Logs lost 

63 

Radiata E 

Radiata R 

Caribbean 

Figure 32. Histogram of recovery based on random data by MGP grade when no more than 5% of the 

potential MGP10 is lost through removal of poor logs. 

The differences are in the percentage of logs removed and the recovery into NS in the 

remnant material. Radiata E had the best results with the largest reduction in NS after drying 

and planing. Radiata R had the worst results with the lowest reduction in NS. This may be 

linked to the good growing conditions for Radiata R, where bigger logs contain a large core of 

large knots, juvenile wood but a large proportion of the periphery contains good quality wood. 

The size of the log facilitates a high recovery of this part. With such good growth conditions 

the “mature” wood is proportionally higher than in the case of tree growing on poorer sites 

where the trees display a slower speed of growth when ageing, leading to smaller log where 

the good part is more difficult to recover. The size of the log is the key: the bigger the log, the 

more variety of wood there is in the log, so rejecting a whole log inevitably also rejects some 

good timber. 

Pre-sorting the cants (including the juvenile core) rather than whole logs, could be a better 

option for the resource, as the wing boards are likely to contain suitable material for structural 

grade products. 

Using best prediction for dry boards NDT technologies 

We have restricted the recovery simulations on dry boards to two different combinations of 

equipment: 

• compression vibration plus WoodEye ® (WE Comp Vib); 

• compression vibration plus WoodEye ® plus Metriguard (WE Comp Vib Metriguard). 

PLS was used with the above combinations of grading technologies to develop predictions for 

the biased test MOR from the grading parameters of each board tested. A separate PLS study

developed predictions of the biased test MOE for the same combinations of grading 

technology. Each of these predictions was then used to assign a grade to the whole board 

based on a threshold. The thresholds were evaluated by trying a number of different 

thresholds with the random position test data, until the pieces that fell into each grade gave 

characteristic values that exceeded the design characteristic value for all grades 

simultaneously. These thresholds could then be used to sort either the biased position test data 

or the random position test data into stress grades. 

Additionally, a composite grading was used in which each of the combinations used both the 

MOE prediction and the MOR prediction, with the requirement that the predictions for any 

board had to exceed both the MOE threshold and the MOR threshold for the grade to be 

admitted to the stress grade. The thresholds were set using the random position test data as 

indicated above. 

Figure 33 and Figure 34 display the correlations between the best grade parameter above and 

MOE (Figure 32) a MOR (Figure 33) for the three resources when applied on biased test data. 

The main observation is the similarity between the correlations on these three different 

resources. This is expressed by a low difference on R 2 associated with a tiny difference on 

both slope and constant of the associated linear equation. 

MoE (MPa) 

25000 

20000 

15000 

10000 

5000 

0 

Radiata E 

Radiata R 

Caribbean 

Linear (Radiata E) 

Linear (Radiata R) 

Linear (Caribbean) 

64 

R² = 0.77 

R² = 0.77 

R² = 0.85 

0 5000 10000 15000 20000 

Grade parameter 

Figure 33. Correlation between MOE and grade parameter from the best combinations of WoodEye ® , 

compression vibration and Metriguard predictors. Biased MOE and MOR threshold combination was 

applied.

MoR (MPa) 

120 

100 

80 

60 

40 

20 

0 

Radiata E 

Radiata R 

Caribbean 

Linear (Radiata E) 

Linear (Radiata R) 

Linear (Caribbean) 

65 

R² = 0.55 

R² = 0.49 

R² = 0.48 

0 2000 4000 6000 8000 10000 12000 14000 16000 18000 20000 

Grade parameter 

Figure 34. Correlation between MOR and grade parameter from the best combinations of WoodEye ® , 

compression vibration and Metriguard predictors. Biased MOE and MOR threshold combination was 

applied. 

Figure 35, Figure 36, and Figure 37 display the coefficient of determination and the 

coefficient of variation for each resource. For a base-line comparison, two combinations and 

Metriguard grading parameters applied to the biased position test data are presented. 

For the Radiata E resource, Figure 35, shows that the R 2 for all grading methods and MOE are 

close to the range of 80% to 90% which is expected to be the maximum attainable within the 

bounds of experimental error. For this resource, most MOE based grading methods were 

demonstrated to perform effectively. There was little difference between the correlations of 

any of the combinations and the Metriguard grading. 

However, Figure 35 also shows that the R 2 for all of the combinations and MOR was better 

than that for the Metriguard grading. As this resource is primarily limited by its MOE 

properties, there may be a large number of MOE measurement devices (including the 

Metriguard) that could return grading as effective as the best of the combination systems. It 

also shows that there is little difference in the COV of bending strength of each of the 

commercial grades and hence all of the plotted grading methods return comparable reliability 

of graded product.

Coefficient of determinationn 

Strength Coefficient of Variation for biased test 

1 

0.9 

0.8 

0.7 

0.6 

0.5 

0.4 

0.3 

0.2 

0.1 

60% 

50% 

40% 

30% 

20% 

10% 

0% 

0 


Correlation coefficient 

66 

MOR WE Comp Vib B 

MOE WE Comp Vib B 

Composite 2 B 

MOR WE Comp Vib Metriguard B 

MOE WE Comp Vib Metriguard B 

Composite 3 B 

Min Metriguard B 

NS MGP10 MGP12 

Grade 

Figure 35. Radiata E resource, coefficient of determination and coefficient of variation for strength 

when grading with the best combinations of WoodEye ® , compression vibration and Metriguard. 

Biased MOE, MOR or combination (composite) of both thresholds was applied. Note: WE Comp Vib 

= 2 and WE Comp Vib Metriguard = 3. 

For Radiata R resource, Figure 36 below shows there is a minor improvement in R 2 with 

MOE using the MOE predictions or the Composite for both combinations. There is a similar 

improvement in R 2 with MOR using the MOR predictions or the Composite for both 

combinations. There was little variation in the COV of MOR for graded material between the 

grading methods with comparable reliability of graded product for all of these grading 

methods.



1 

0.9 

0.8 

0.7 

0.6 

0.5 

0.4 

0.3 

0.2 

0.1 

60% 

50% 

40% 

30% 

20% 

10% 

0% 

0 



67 



Composite 2 B 



Composite 3 B 


NS MGP10 MGP12 

Grade 

Figure 36. Radiata R resource, coefficient of determination and coefficient of variation for strength 



= 2 and WE Comp Vib Metriguard = 3. 

For Caribbean resource the use of a combination of grading equipment or predictors 

significantly increased the R 2 of both MOE and MOR. The R 2 with MOE was increased most 

using the MOE prediction and the R 2 with MOR was increased most using the MOR 

prediction. These and the composite grading returned higher R 2 than the base-line Metriguard 

grading for both MOE and MOR as shown in Figure 37.

The COV of the MOR for each of the MGP12 and MGP15 grades, also shown in Figure 37, 

showed a significant reduction below its Metriguard base-line value with the application of 

new technology. This resource offered the most potential for improvement of grading using a 

combination of grading technologies. 



1 

0.9 

0.8 

0.7 

0.6 

0.5 

0.4 

0.3 

0.2 

0.1 

60% 

50% 

40% 

30% 

20% 

10% 

0 



68 



Composite 2 B 



Composite 3 B 


0% 

NS MGP10 MGP12 MGP15 

Grade 

Figure 37. Caribbean resource, coefficient of determination and coefficient of variation for strength 



= 2 and WE Comp Vib Metriguard = 3.

Recoveries based on random test 

70% 

60% 

50% 

40% 

30% 

20% 

10% 

0% 


Grade 

Figure 40 display the recoveries from the application of the same grade thresholds illustrated above to 

the full suite of boards represented by the random data. 

69 

MOR WE Comp Vib R 

MOE WE Comp Vib R 

Composite 2 R 

MOR WE Comp Vib Metriguard R 

MOE WE Comp Vib Metriguard R 

Composite 3 R 

Min Metriguard R 

The important thing to focus on in these graphs is the recovery to non-structural (NS). We 

want to minimise this as this is timber that loses money for the mill. The tendencies observed 

on the quality regression indicators when applying the best predictions on biased test data are 

confirmed on the recoveries while applying on random test data, namely: 

• Radiata E resource: there is no improvement observed when using the best 

combinations of WoodEye ® , compression vibration and Metriguard compared to the 

use of Metriguard alone. 

• Radiata R resource: the recovery increase observed on MGP10 grade is somewhat 

offset by an increase of non structural products and a recovery decrease in MGP 12 

grade. 

• Caribbean resource: a significant recovery increase on MGP10 and MGP15 grades 

associated with an important decrease of non structural products. Despite an absolute 

decrease of 15% on MGP12 recovery, the 10% absolute decrease of non structural 

products associated with a 20% absolute increase on MGP10 recovery represents 

certainly a key advantage for this resource.

70% 

60% 

st 

te 

m 

o 50% 

d 

n 

ra 

n 40% 

o 

d 

b 

a 

se 

30% 

s 

v 

e 

rie 

20% 

e 

c 

o 

R 

10% 

0% 


Grade 

Figure 38. Radiata E recoveries from calibrations developed with biased test data for each resource 

and for different grading parameters when using static bending random data. Biased MOE, MOR or 

combination (composite) of both thresholds was applied. Note: WE Comp Vib = 2 and WE Comp Vib 

Metriguard = 3. 

70 



Composite 2 R 


Radiata E MOE WE Comp Vib Metriguard R 

Composite 3 R 

Min Metriguard R


70% 

60% 

50% 

40% 

30% 

20% 

10% 

0% 


Grade 

Figure 39. Radiata R recoveries from calibrations developed with biased test data for each resource 



Metriguard = 3. 


70% 

60% 

50% 

40% 

30% 

20% 

10% 

0% 

71 



Composite 2 R 



Composite 3 R 




Composite 2 R 



Composite 3 R 



Grade 

Figure 40. Caribbean recoveries from calibrations developed with biased test data for each resource 



Metriguard = 3.

Discussion and conclusions 

The capability of a grading system to predict accurately and reliably the reference parameters 

of a product, namely the board’s actual stiffness and strength, depends mainly on how well 

the prediction agrees with: 

• the reference parameters (indicated by R 2 , the coefficient of determination); 

• the measurement error of the predictor parameter(s) (indicated by COV, the 

coefficient of variation); and 

• the capability of the system to accommodate resource variability (indicated by its 

prediction robustness and the level of recovery compared to the actual grading based 

on destructive tests). 

If the regression analysis is based on measurements made in the same conditions and the same 

apparatus that is used in grading machines, the effect of the measurement error and COV is 

already included in the R 2 value directly. This was the case for the in-line equipment tested. 

For the resonance methods made in laboratory conditions with the Bing ® equipment, the effect 

of measurement error should be considered separately, when evaluating its effectiveness. 

However resonance methods have proven to be very robust and accurate due to their 

metrological 1 simplicity and fundamental principles. There is a large quantity of publications 

based on theoretical mechanics work and experimental results which demonstrate the 

capability to predict stiffness and strength parameters of wood and other materials. In this 

study the comparison of results obtained between in-line equipment (Thumper and 

Metriguard) and a laboratory system (Bing ® ) demonstrated the high degree of closeness 

between in-line and laboratory measurements on one hand, and the quality of the prediction of 

the reference static bending MOE and MOR on the other hand. Theoretical, flexion should 

provide the best results due to the parallels to static bending, that is the configuration implied 

flexure stress. However, due to metrological constraints, low frequencies and signal duration, 

plus boundary conditions, compression and flexion vibration provide the same level of 

prediction. 

The sample size for each resource (approximately five hundred boards per resource) was large 

enough to make the results reliable. The measurements of the various parameters of this study 

were performed at different times and sometimes under dissimilar conditions. Despite the 

effort to keep these conditions as homogenous as possible some variations may persist. As a 

consequence, the results presented here should not be used as definitive when the difference 

between the R 2 is typically less than 0.02. 

As was found in the European Combigrade projects’ (Hanhijärvi et al. 2008 and 2005) results 

for dry boards, this study demonstrates that the best single parameter predictors of MOE and 

MOR are the stiffness related measurements derived by either direct mechanical test 

(Metriguard) or vibration method. These two approaches offer a similar level of prediction 

which the R 2 ranges from 0.66 to 0.85 for MOE and 0.35 to 0.53 for MOR, depending on the 

resource. Importantly because the Metriguard provides both average and local board 

measurements it shows better performance on strength limited or defect affected resources. 

On clearly stiffness-limited resource, vibration seems to provide a slightly better prediction 

than Metriguard. On this type of resource combining stiffness parameters with local 

measurements, knot or local stiffness, does not improve the prediction significantly. 

1 relating to metrology. Metrology is the scientific study of measurement. 

72

Reference static bending parameters prediction for strength limited, or defect affected 

resources, can be notably improved by combining stiffness parameters with local information 

and an R 2 improvement magnitude up to 0.1 for MOE and 0.2 for MOR can be expected. 

The choice of the statistical approach to achieve these improvements is determinant. The 

application of classical MLR on several predictors proved to be effective, in particular for 

MOR predictions. However the usage of a modern statistics approach (chemometrics tools) 

such as PLS proves to be more efficient for improving the level of prediction. This is due to 

the problem of multi-collinearity in MLR which impaired the efficient exploitation of the 

information contained in all the predictors available for a given combination of grading 

systems. As the development of robust calibration is based on training samples (samples 

specifically selected to develop the prediction equation) and application of complex 

mathematics, the robustness and accuracy of the prediction requires a sound knowledge and a 

disciplined approach. 

Obviously statistical analysis alone is only a criterion, even if decisive, when considering the 

fitness of a grading system to a certain application. The price of the system, its maintenance 

cost, its production line adaptation, versatility, reliability, simplicity, etc. are all important 

factors. 

Moreover, the reference method accuracy is not perfect. The impact of the experimental error 

on the reference’s actual parameters leads to a maximum attainable level of prediction within 

the bounds of error. When the quality of the prediction comes closer to this threshold, any 

improvement requires much more effort and consequently investment. The balance between 

grade selection accuracy and investment depend on each processor’s strategy. 

The consequence of the application of combined grading systems on grade recoveries has to 

be analysed carefully since it is clearly resource dependent. The trend observed shows that 

there is no advantage to using combined grading systems for a stiffness-limited resource. The 

advantage for a strength limited resource has to be weighed with other processing impacts. 

The Caribbean resource offers the most promising improvement potential using a 

combination of new technologies. 

The use of resonance method for MOE and MOR prediction on green boards provides slightly 

less successful results when compared to dry boards. Nevertheless, the R 2 decrease is only 0.1 

to 0.15 depending on the resource. This result is a real opportunity for processors to try to 

apply an early detection of the weakest boards in order to relieve the cost of non-value added 

further processing. Moreover the potential use of optical scanners or equivalent equipment 

allowing the detection of grain deviation and knot characteristics on green boards promises to 

increase quite significantly the MOR prediction. It represents an avenue to improve the 

efficiency of the mill. 

Log segregation through vibration method (compression) is also resource dependent. From 

the results of this study, it is clear that the stiffness-limited resource should have the highest 

gain in recovery, through removal of the lowest stiffness logs. Therefore, significant cost 

savings could be achieved by processing these into non-structural products, with minimal 

impact on final graded product recovery. The value gain from log stiffness sorting is less for 

Caribbean pine and requires further assessment. For strength limited resources, the 

consequences on recovery are insignificant. 

73

Recommendations 

All the results obtained from this study should be validated with full in-mill simulations. This 

could be achieved using calibrations developed during this project and applied to independent 

samples in sufficient numbers of replicates covering the range of variability of material 

handled by each processor. 

There is room for improvement on each system used, for example the use of an optical system 

could be tuned for the resource’s grade limiting features. The WoodEye ® system used in this 

study was tuned for appearance grading of hoop pine and it was outside the scope of the 

project to re-tune the equipment for structural grading purposes. Moreover, the raw data 

variables extracted from WoodEye ® weren’t refined to provide the most appropriate 

characterization of the material tested. Further discussion with qualified optical scanning 

system engineers should provide optimized extraction for structural grading purposes. This 

will involve the development of clear definitions of the optical characteristics that impact 

MOE and MOR for each resource. 

Vibration method can be improved through: 

• installation of devices at strategic locations in the mill (i.e. cants); 

• refining data extracted, e.g. using current dataset, further analysis to improve 

algorithm extraction; 

• using flexure with different boundary conditions, e.g. testing local span within boards. 

At green board level, an optical scanning system coupled with a vibration system, has proved 

to be nearly as efficient as the same combination on dry boards. Depending on the processor’s 

strategy, further investigation may provide an avenue for efficient green mill grading 

technologies. 

A larger sample of logs should be tested to verify the potential for effective log segregation. 

An error analysis on the reference method for MOE/MOR measurement should be performed 

in order to assess the maximal level of prediction achievable. 

Locating vibration devices at strategic locations in the mill may provide a cost-effective 

alternative to investment in higher cost capital items installed near to the end of the process 

line. 

A further project could be designed to continue mining the existing dataset. For example, 

testing the consequence of grading at stages through the flow of the boards. 

74

References 

Baradit, E., Aedo R. and Correa J. (2006). Knots detection in wood using microwaves, Wood 

Science and Technology 40 (pp118-123). 

Bendtsen B.A. (1978) Properties of wood from improved and intensively managed trees. 

Bootle K.R., (2005). WOOD IN AUSTRALIA. TYPES, PROPERTIES AND USES. 2nd Ed. 

McGraw-Hill Book Company, Sydney, Australia. 

Brancheriau, L. (2002). Expertise mécanique des sciages par analyses des vibrations dans le 

domaine acoustique, Université de la méditerranée - Aix Marseille II, Université de la 

méditerranée - Aix Marseille II, Marseille (p 278). 

Brancheriau, L. (2007). WISIS - Wood In Situ Inspection- Instructions for use Version 1.0. 

CIRAD. 

Brancheriau, L. and Bailleres, H. (2002). Natural vibration analysis of clear wooden beams: a 

theoretical review. Wood Science and Technology 36 (pp 346-365). Springer-Verlag. 

Brancheriau, L. and Bailleres, H. (2003). Use of Partial Least Squares Method with Acoustic 

Vibration Spectra as a New grading Technique for Structural Timber. Holzforschung Vol. 57, 

No. 6 (pp 644-652). 

Bucur, V. (2003). NONDESTRUCTIVE CHARACTERIZATION AND IMAGING OF 

WOOD. Springer-Verlag, Berlin, Germany. 

Duff, G. (2005). Technology for delivering high quality graded softwood product- practical 

applications. Gottstein Fellowship report. 

Grabianowski, M., B. Manley, and J.C.F. Walker. 2006. Acoustic measurements on standing 

trees, logs and green lumber. Wood Sci. Technol. 40(3):205-216. 

Halabe, U.B., Bidigalu G.M., GangaRao H.V.S. and Ross R.J. (1995). Nondestructive 

Evaluation of Green Wood Using Stress Wave and Transverse Vibration Techniques. Vol. 55, 

No. 9. 

Hanhijärvi, A. and Ranta-Maunus, A (2008). Development of strength grading of timber 

using combined measurement techniques. Report of the Combigrade project – phase 2. VTT 

Publications 686. 

Hanhijärvi, A., Ranta-Maunus, A and Turk, G. (2005). Potential of strength grading of timber 

with combined measurement techniques. Report of the Combigrade project – phase 1. VTT 

Publications 568. 

Harris J.M. (1991). PROPERTIES AND USES OF NEW ZEALAND RADIATA PINE. 

Volume one - wood properties, Forest Research Institute, Rotorua, New Zealand. 

Hopewell, G. (Ed.) 2006. CONSTRUCTION TIMBERS IN QUEENSLAND: PROPERTIES 

AND SPECIFICATIONS FOR SATISFACTORY PERFORMANCE OF CONSTRUCTION 

75

TIMBERS IN QUEENSLAND, CLASS 1 AND CLASS 10 BUILDINGS (BOOKS 1 AND 

2). Department of Primary Industries and Fisheries, Brisbane. 

Ilic, J., Northway, R. and Pongracic, S. (2003). Juvenile Wood Characteristics, Effects and 

Identification. Literature Review. FWPRDC Project No. PN02.1907. 

Kininmonth, J.A., Whitehouse, L.J. (1991). Properties and Uses of New Zealand Radiata 

Pine, Vol. 1. Forest Research Institute, Rotorua, New Zealand, 

Krautkrämer, J. K.H. (1986). Werkstoffprüfung mit Ultraschall, Springer Verlag, Berlin. 

Krzosek, S. (2003). Machine stress-grading of timber: current status and prospects of 

development, Festigkeitssortierung des Bauholzes mit Maschinen - Ist-Stand und 

Entwicklungsperspektiven. 

Lamb A.F.A., (1973). Pinus caribaea FAST GROWING TREES OF THE LOWLAND 

TROPICS Volume 1. N°6. University of Oxford, England (254 pp). 

Leicester, R.H. (2004). Grading for structural timber. Progress in Structural engineering and 

Materials. Vol. 6 (pp 69-78). 

Leicester, R.H., Breitinger, H.O. and Fordham, H.F. (1996). Equivalence of in-grade testing 

standards. CIB/W18A- Timber Structures, Paper 29-6-2, Bordeaux, France. 

Ross, R.J. and Pellerin, R.F. (1994). Non-destructive testing for assessing wood members in 

structures- a review. (40 pp). Rev. Ed. Forest Products Laboratory USDA Forest Service, 

Madison USA. 

Ross, R.J., Brashaw B.K. and Wang X.P. (2006). Structural condition assessment of inservice 

wood. Forest Products Journal 56, 4. 

Ross, R.J., Zerbe, J.I., Wang, X.P., Green D.W. and Pellerin R.F. (2005). Stress wave 

nondestructive evaluation of Douglas-fir peeler cores. Forest Products Journal 55 (pp 90-94). 

Sandoz, J.L. (1989). Grading of construction timber by ultrasound. Wood Science and 

Technology 23 (pp 95-108). 

Schimleck, L.R., Evans, R., Ilic, J. and Matheson A.C. (2002). Estimation of wood stiffness of 

increment cores by near-infrared spectroscopy. Canadian Journal of Forest Research 32 (pp 

129-135). 

Siemon, G. (1981). Effect of CCA-preservative treatment on bending strength of small clear 

specimens of high-temperature dried and air-dried Caribbean pine. Technical Paper No. 27. (4 

pp). Department of Forestry, Queensland. 

Smith, R.G.B., Palmer, G. Davies, M. And Muneri, A (2003). A method enabling the 

reconstruction of internal features of logs from sawn lumber: the log end template. Forest 

Products Journal Volume 53 No. 11/12 (pp 95-98). 

Steiger, R. (1996). Mechanische Eigenschaften von Schweizer Fichten-Bauholz bei Biege-, 

Zug-, Druck- und kombinierter M/N-Belastung Sortierung von Rund- und Schnittholz mittels 

76

Ultraschall, VII, 168S. Institut für Baustatik und Konstruktion (IBK), Eidgenössische 

Technische Hochschule Zürich (ETH), Zürich. 

Tsehaye, A.., Buchanan, A.H. and Walker, J.C.F. (2000). Sorting of logs using acoustics. 

Wood Science and Technology 34 (pp 337-344). 

Tsuchikawa, S. (2007). A Review of Recent Near Infrared Research for Wood and Paper. 

Applied Spectroscopy Reviews 42 (1) (pp 43–71). 

Wang, X.P., Ross, R.J., Green, D.W., Brasha, B., Englund, K. and Wolcott, M. (2003). Stress 

wave sorting of red maple logs for structural quality. Wood Science and Technology 37 (pp 

531-537). 

Yeh, T.F., Chang, H.M. and Kadla, J.F. (2004). Rapid prediction of solid wood lignin content 

using transmittance near-infrared spectroscopy. Journal of Agricultural and Food Chemistry 

52 (pp 1435-1439). 

Zobel, B.J. and Buijtenen J.P.V (1989). WOOD VARIATION: ITS CAUSES AND 

CONTROL. Springer-Verlag, Berlin, Germany. 

Internet references 

Web 1. http://www.weeds.org.au/cgi-bin/weedident.cgi?tpl=plant.tpl&ibra=all&card=E41# 

Visited in April 2007 

Web 2. http://www.statsoft.com/textbook 

Visited 31.07.09 

Web 3. http://www2.dpi.qld.gov.au/hardwoodsqld/13322.html 

Visited in August 2008. 

Web 4. http://www.metriguard.com/ 

Visited in August 2008 

Web 5. http://www.newnes-mcgehee.com/splash.htm 

Visited in August 2008 

Standards 

AS/NZS 1748:2003 - Timber-Stress-graded-product requirements for mechanically stressgraded 

timber. 

EN 408 (1995). European Standard 408, Timber structures- Structural timber and glued 

laminated timber- Determination of some physical and mechanical properties. Brussels. 

EN 14081-1 (2003). European Standard E. Timber structures- Strength graded structural 

timber with rectangular cross section- Part 1: General requirements. Brussels. 

AS 1720.1 (1997) Australian Standard. Timber structures, Part 1: Design methods. 

77

AS/NZS 4063 (1992). Australian/New Zealand Standard 4063. Timber−Stress-graded−Ingrade−strength 

and stiffness evaluation. 

Personal communication 

Boughton, G. (2009). pers. comm. Director, TimberED, Bunbury Western Australia. 

Nikles, G. Dr. (2009). Research Associate, Queensland Primary Industries and Fisheries. 

78

Acknowledgements 

The successful conduct of this project was made possible through the enthusiastic support 

from management and staff of the following organisations: 

• Carter Holt Harvey, Myrtleford Victoria 

• Carter Holt Harvey, Caboolture Queensland (formerly Weyerhaueser) 

• Wespine, Bunbury Western Australia 

• Hyne Timber, Tumbarumba New South Wales and Tuan Queensland 

• Queensland Primary Industries and Fisheries, Brisbane Queensland 

• TimberED, Perth Western Australia 

• CIRAD Xylometry, Montpellier France 

79

Researcher’s Disclaimer 

© The State of Queensland, Primary Industries and Fisheries, 2009. 

Except as permitted by the Copyright Act 1968, no part of the work may in any form or by 

any electronic, mechanical, photocopying, recording, or any other means be reproduced, 

stored in a retrieval system or be broadcast or transmitted without the prior written permission 

of DPI&F. The information contained herein is subject to change without notice. The 

copyright owner shall not be liable for technical or other errors or omissions contained herein. 

The reader/user accepts all risks and responsibility for losses, damages, costs and other 

consequences resulting directly or indirectly from using this information. 

80

Appendix 1. Resonance method extracted signal descriptors 

SCGravity 

Spectral centre of gravity divided by the fundamental frequency (f1) in %: 

SCGf1 = SCG ./ f1 .* 100; 

SBandwidth 

with f(N) = Fmax range 

Spectral extent divided the fundamental frequency (f1) in %: 

Sbandf1 = Sband ./ f1 .* 100; 

SSlope 

Spectral slope divided the fundamental frequency (f1) in %: 

Sslopef1 = Sslope ./ f1 .* 100; 

IF 

Inharmonicity factor: 

ME 

Average of the first three dynamic MOE 

MOE_n 

Dynamic MOE associated with fn 

FacQ_n 

Quality factor (inverse internal damping or friction) associated with fn: 

81

Pow_n 

such as 

Mean power in frequency sub-band (between 0 and f1; f1 and f2...): 

MSNrjRatioF_n 

Sub-band energy ratio (between the resonance frequencies) : 

PowF_n 

= Sub-band energy / Total range energy 

Mean power of the sub-band (resonance shoulder defined by a pass band of -20dB) : 

MSRatioF_n 

Sub-band energy ratio (resonance shoulder defined by a pass band of -20dB) 

= sub-band energy / total range energy 

82

CTR 

Appendix 2. WoodEye profile descriptions. 

CTR_AC_M 

CTR_AC_5 

CTR_BD_5 

CTR_trgl_BD_5 

CTR_trgl_AC_M 

CTR_trgl_AC_3 

CTR_trgl_AC_5 

______________ 

KAR 

KAR_ext_M 

KAR_ext_99 

KAR_ext_98 

KAR_int_M 

KAR_int_99 

KAR_trgl_M 

KAR_trgl_99 

Material inertia ratio = calculated inertia / beam inertia (b.h³/12) 

The inertia is calculated from the sum of inertias according to Parallel Axis Theorem 

(also known as Huygens-Steiner theorem) 

Profile is calculated between the external loading points 

Profile is weighted by the bending moment (unit maximum moment) 

Profile is equal to the product of the profiles from the 4 beam sides. 

Sides A&C, rectangular window, mean value 

Sides A&C, rectangular window, 5% fractile (equivalent to a minimum) 

Ratio defect surface/total surface; B&D edges, rectangular window, 5% fractile 

B&D edges, triangular window, 5% fractile 

Sides A&C, triangular window, mean value 

Sides A&C, triangular window, 3% fractile 

Sides A&C, triangular window, 5% fractile 

____________________________________________________________ 

Perimeter ratio = calculated perimeter / beam perimeter (2.b.h) 

The perimeter is associated to defect height (sum of the heights) 

Profile is calculated between the external loading points 

Profile is equal to the sum of the profiles from the 4 beam sides. 

A&B&C&D, 1/3 external height for A&C), rectangular window, mean value 

99% fractile (equivalent to a maximum) 

98% fractile 

A&B&C&D, 1/3 internal height for A&C), rectangular window, mean value 

99% fractile 

A&B&C&D, triangular window, mean value 

99% fractile 

Notes: A and C are the faces and B and D are the board edges. 

83

Appendix 3. Biased and random testing protocol. 


• Identify the limiting feature (that is, the largest strength limiting defect) to provide the 

biased sample location. 

• Mark the 1.8 m length so that this point lies within the middle third of the specimen. 

• Using a random number calculator, provide the distance from the board end for 

random sample datum point. 

• If this position is situated so that a second 1.8 m long test piece can be created without 

overlapping the biased specimen, it is marked ‘true random’. 

• Otherwise a sample is cut directly next to the biased one and marked ‘false random’. 

• When the biased and the random samples are situated so that both conditions were 

satisfied, the specimen is marked with ‘biased and random’. 

• The specimens called true and false random are both considered as random samples. 

• The results obtained from the ‘biased and random’ samples are taken into account in 

random as well as biased analyses. 

This sampling methodology, whereby the biased sampling had priority and a random sample 

was only available if located outside the biased sample region, provided a random sample set 

of approximately 40% of the whole sub-sample. 

Radiata R and Caribbean resources 

• Using a random number calculator, provide the distance from the numbered board end 

for the random sample datum point. 

• A 1800 mm test specimen template with the centre third demarcated is placed at this 

datum to locate random specimen location. 

• Check to see if the maximum strength reducing defect is located in the centre third of 

the possible test specimen. 

• If yes then mark the test specimen for docking and retain the board number on the 

specimen with the letters “rb” for “random and biased”. 

Random length measured from the end with the full board identification code 

End with the full board identification code 

Maximum strength reducing defect 

• If no then locate the maximum strength reducing defect in the board and position the 

test specimen template so that the defect is in the middle third of the test span. 

• Mark the test specimen ends for docking and retain the board number on the test 

specimen with the letters “b” after the sample number (“b” indicates a biased sample). 

• If the random sample doesn’t overlap the biased sample, two samples can be cut. 

84




• If the biased and random samples overlap slightly, the combined length is docked and 

the biased test section tested first, followed by the random section length. 




• If the overlap is too large only one test can be done the piece with the priority has to 

be selected (random or biased). In this case only one specimen will result. 




This protocol would ensure that for approximately 60% of boards, both random and biased 

test specimens will be available. 

85

MOE and MOR assessment technologies for improving graded ...

Create successful ePaper yourself

Delete template?

Save as template?